WPS6244


Policy Research Working Paper                        6244




        Heterogeneity in Subjective Wellbeing
      An Application to Occupational Allocation in Africa

                                    Paolo Falco
                                William F. Maloney
                                    Bob Rijkers
                                 Mauricio Sarrias




The World Bank
Development Research Group
Macroeconomics and Growth Team
 &
Middle East and Northern Africa Region
Office of the Chief Economist
October 2012
Policy Research Working Paper 6244


  Abstract
  Using an extraordinarily rich panel dataset from Ghana,                           distribution of parameters predicting satisfaction reveal
  this paper explores the nature of self-employment and                             substantial heterogeneity in subjective well-being within
  informality in developing countries through the analysis                          sectors that conventional fixed parameter models, such as
  of self-reported happiness with work and life. Subjective                         standard ordered probit models, cannot detect: Whatever
  job satisfaction measures allow assessment of the relative                        the average satisfaction premium in a sector, all job
  desirability of different jobs in ways that, conditional                          categories contain both relatively happy and disgruntled
  wage comparisons cannot. By exploiting recent advances                            workers. Specifically, roughly 67, 50, 40 and 59 percent
  in mixed (random parameter) ordered probit models,                                prefer being a small-firm employer, sole proprietor,
  the distribution of subjective well-being across sectors                          informal salaried, civic worker respectively, than formal
  of employment is quantified. There is little evidence for                         work. Hence, there is a high degree of overlap in the
  the overall inferiority of the small firm informal sector:                        distribution of satisfaction across sectors. The results
  there is not a robust average satisfaction premium for                            are robust to the inclusion of fixed effects and alternate
  formal work vs. self-employment or informal salaried                              measures of satisfaction. Job characteristics, self-perceived
  work, and owners of informal firms that employ others                             autonomy and experimentally elicited measures of
  are on average significantly happier than workers in                              attitudes toward risk do not appear to explain these
  the formal private sector. Moreover, the estimated                                distributional patterns.




  This paper is a product of the Macroeconomics and Growth Team, Development Research Group; and the Office of the
  Chief Economist, Middle East and Northern Africa Region. It is part of a larger effort by the World Bank to provide
  open access to its research and make a contribution to development policy discussions around the world. Policy Research
  Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at wmaloney@
  worldbank.org and brijkers@worldbank.org.




         The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development
         issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the
         names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those
         of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and
         its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.


                                                       Produced by the Research Support Team
  Heterogeneity in Subjective Wellbeing: An Application
          to Occupational Allocation in Africa∗
   Paolo Falco†          William F. Maloney‡             Bob Rijkers§          Mauricio Sarrias¶
                                             October 9, 2012




Keywords: Subjective Wellbeing, Mixed Ordered Probit, Self-employment, Informality, Developing Country
Labor Markets, Africa
JEL codes: C35, J2, J3, J41, L26, I32, 017
Sector board: EPOL

   ∗
     The views expressed here are those of the authors and do not necessarily represent the views of the
World Bank, its Executive Board or member countries. We are grateful to Kathleen Beegle, Caroline Freund,
Francis Teal, Carolina Sanchez Paramo, Mary Hallward-Driemeier and attendants of the 2011 CSAE and
NEUDC conferences for valuable comments and to the World Bank’s RSB Committee for funding. All errors
are our own.
   †
     Centre for the Study of African Economies, Department of Economics, Oxford University
   ‡
     Development Economics Research Group, The World Bank, and Universidad de los Andes, Bogota,
Colombia
   §
     Chief Economist Office for the Middle East and North Africa, The World Bank
   ¶
     Universidad Catolica del Norte, Antofagasta, Chile
                                                    1
1       Introduction
Subjective measures of job and life satisfaction have proven good proxies for both job quality
and wellbeing (see e.g., Oswald, 1997; Frey and Stutzer, 2002), important determinants of
economic behavior, and powerful predictors of job tenure (e.g Freeman, 1978; Akerlof et al.,
1988), productivity (Oswald et al., 2008) and future earnings (e.g Wright and Staw, 1999).
They are particularly useful tools in assessing the relative desirability of different employ-
ment sectors since the weights needed to combine various observable job characteristics
into a unidimensional metric are typically not known and may vary across individuals with
different preferences (Clark and Senik, 2010a),1 and because some of the most important
job attributes may be unobservable.              For example, settling the debate over whether
self-employment is a desirable option relative to salaried employment has been complicated
by the difficulty of measuring and weighting such factors as the appropriate risk premium,
aversion to hierarchy, or the value of flexibility.             Subjective indicators mitigate these
problems by virtue of being comprehensive and relying on individuals’ own weighting of
various attributes, and consistently suggest the existence of a self-employment satisfaction
premium, both in developed and developing countries (Blanchflower and Oswald, 1998;
Blanchflower, 2000; Idson, 1990; Benz and Frey, 2008a,b).


However, the central tendency of the satisfaction premium alone may be insufficient to cap-
ture the richness of the processes that determine sectoral allocation and subjective wellbeing.
For instance, Evans and Leighton (1989) also argue for the presence in self-employment in
the U.S. of “mis�ts cast off from wage work� who are likely to have experienced a fall in
satisfaction in the transition from wage to self-employment.2 It is thus possible that despite
a positive average premium in self-employment, for a large share of individuals in the sector
the premium is negative. More generally, the notion of latent heterogeneity underpins many
matching models of the labor market, and is often appealed to in explaining why agents
    1
      Relatedly, in explaining why objective and subjective poverty measures diverge signi�cantly in Russia
Ravallion and Lokshin (2002) suggest that the weights assigned to different elements used to construct
objective poverty lines might be inappropriate and that the low dimensionality of the objective measure
of poverty misses key dimensions of percieved poverty
    2
      “The disadvantage theory which views entrepreneurs as mis�ts cast off from wage work is consistent
with many of our �ndings. People who switch from wage work to self-employment tend to be people who were
receiving relatively low wages, who have changed jobs frequently, and who experienced relatively frequent or
long spells of unemployment as wage workers.� Evans and Leighton (1989) p. 532
                                                     2
with identical observable characteristics exhibit differential responses to common shocks,
such as policy changes. A crucial feature of these types of models is that individuals differ
in the amount of utility they derive from being in a particular job. Such differences may
arise from differences in the preferences of the worker, or the characteristics of the job.
Hence, exploring the latent heterogeneity of satisfaction within sectors is important for our
understanding of labor markets.


This paper undertakes such an exploration by using recent advances in stochastic random
parameter (mixed) discrete choice models to provide a more complete description of the
distribution of subjective welfare across employment sectors. We do so in the developing
country context where the role and implications of the extensive self-employed and small �rm
sector – broadly termed the informal sector – have been intensely debated for decades. While
the advanced country literature stresses the desirability of independence and being one’s own
boss that self-employment and the small �rm sector offers, the developing country literature
has tended to conclude from the attendant lack of social protection, and the association
with poverty more generally, that such jobs are the inferior part of a highly segmented
labor market.3 However, demonstrating segmentation requires showing that, at the margin,
utility is not equated across sectors and the conditional wage premia commonly estimated
as proxies may instead reflect preference heterogeneity and compensating differentials (e.g.
for risk, independence, taxes avoided, the perceived value of bene�ts, or training). Hence,
as with the advanced country literature on self-employment, establishing the distribution of
subjective measures of satisfaction offers more potential for characterizing the informal sector.

   3
     Most evidence for segmentation relies on the observation that there is a sizeable formal sector
wage premium; larger �rms pay workers with otherwise similar observable characteristics more (see
S¨oderbom et al., 2006); and sorting is a key determinant of differences in labor income(Fafchamps et al.,
2009). Yet evidence based on longitudinal data on labor market transitions from Latin America (see
e.g. Maloney, 1999; Gong and Van Soest, 2002; Gong et al., 2004; Bosch and Maloney, 2006, 2010) suggests
that characterizing self-employment as inferior to wage employment may be inappropriate since for many
workers self-employment is a desirable alternative to formal sector employment, offering more flexibility and
better pay. Although studies of this type are less prevalent in Africa, there are signs that African labor
markets may not be highly segmented. To start with, average wages have been surprisingly responsive to
unemployment rates (Kingdon et al., 2005). Secondly, while earnings vary systematically across sectors and
are higher in formal wage employment, there is tremendous heterogeneity in returns within wage- and self-
employment (Falco et al., 2010); many of the self-employed earn more than comparable individuals in wage
jobs. Moreover, Launov and Gunther (2012) test the dualistic labor market hypothesis by means of a mixture
model that allows for endogenous segment selection using data from Cote d’Ivoire and reject it in favor of the
Cunningham and Maloney (2001) and Fields (2005) views that argue that the informal sector has its own
internal dualism.
                                                      3
To this end, we exploit the Ghana Urban Household Panel Survey (GUHPS), a data set
unique in the African context and one of the very few longitudinal datasets in either the
developing or advanced world containing several measures of subjective satisfaction with
work, life and �nances. By income, Ghana is broadly representative of sub-Saharan Africa
and offers insight into low and low middle income countries more generally.4 The GUHPS
permits us to study job satisfaction across sectors, potentially controlling for unobserved
heterogeneity, and thereby to better characterize developing country labor markets and
assess how well the developed country literature generalizes to them. Using conventional
ordered categorical data models, we �rst test for the existence of self-employment and
small �rm satisfaction premia. Since studies of job satisfaction in developing countries are
scarce (see Clark and Senik, 2010a, for a review of the literature) it is unknown whether
this �nding translates to poor countries where such jobs often constitute the vast majority
of employment, sometimes accounting for 70-80% of the labor force (see Kingdon et al.,
2005).5 The few studies of job satisfaction in developing countries that we are aware of tend
to focus on a select subset of workers (Mulinge and Mueller, 1998), and rely exclusively
                                                                            oo et al., 2010;
on cross-sectional datasets (Pratap and Quintin, 2006; Perry et al., 2007; B´
   es et al., 2008). In particular, we test whether the relative satisfaction premium for
Pag´
self-employment and informal salaried employment are more consistent with the competitive
view emerging from the OECD literature, or a dualistic conceptualization of the informal
sector: whether there is a positive satisfaction premium associated with formal jobs that
would signal that markets are not competitive, as well as identifying which characteristics
may be associated with it.


We then confront heterogeneity in conditional satisfaction head on by means of mixed
(random parameter) ordered probit models. These models extend the mixed logit model to
ordinal responses and can be seen as a generalization of the latent class techniques used by
Clark (2004). The mixed model not only enables us to identify the mean premium associated
   4
      Ghana’s 2010 GNI/capita of $US 1,230 is roughly the average for sub-Saharan Africa ($US 1,176).
Though reliable representative labor market data are fairly sparse, our estimates of the share of self-
employment is broadly similar to that for sub-Saharan Africa.
    5
      The ILO (2002) estimates that informal sector self-employment alone accounts for 53% of all non-
agricultural employment in Sub-Saharan Africa and about a third of all non-agricultural employment in
North Africa.
                                                  4
with being in a particular sector, but also to assess whether, and to what extent, conditional
job satisfaction varies within those sectors. We document signi�cant heterogeneity in job
satisfaction across sectors and are able to quantify what fraction of workers are voluntary
or involuntary in each sector, as well as the degree of dispersion of satisfaction. Further,
the panel dimension allows us to estimate a random effect mixed ordered probit model.
This allows us to control for individual-speci�c predispositions that affect self-reported
satisfaction across all sectors. The results also shed light on the empirical relevance of
multisector labor market models (see e.g. Fields (2005)), including those that stipulate that
the informal sector has its own internal dualism and should therefore be characterized by
heterogeneity in self-reported job satisfaction.


Finally, we conduct a number of robustness checks and examine a number of previously
underexplored explanations for the distribution of job satisfaction across sectors. First,
we assess the robustness of our results using the �xed effects logit model proposed by
Ferrer-i Carbonell and Frijters (2004).         Second, we test for robustness of the results
to alternate measures of wellbeing, notably life and �nancial satisfaction and examine
whether differences in job satisfaction are driven by tradeoffs among these (as suggested
by Benz and Frey (2008a)).         Third, as autonomy has been argued to be an important
reason for the higher job satisfaction levels reported by the self-employed in a wide variety
of countries (Benz and Frey, 2008a), we examine the importance of a feeling of “control�
in driving the observed distribution. Fourth, we assess the robustness of our results to
controlling for a richer set of job (as opposed to worker) characteristics. Finally, using an
experimentally elicited measure of risk aversion, we examine to what extent differences in
satisfaction across jobs are the result of differences in workers’ risk tolerance focusing in par-
ticular on the question as to whether less risk-averse individuals enjoy self-employment more.6


The remainder of this paper is organized as follows. Section 2 presents our econometric
strategy and formulates the mixed ordered probit model. Section 3 describes our data,
explains how subjective wellbeing is measured, and presents descriptive statistics. Section 4
presents our baseline results using speci�cations that are standard in the literature. Section 5
   6
     Note that self-employment need not be more risky; for example, wage employees may face a substantial
risk of being �red.
                                                   5
presents estimates using our mixed ordered estimators. Section 6 presents robustness checks
and explores a number of salient explanations for the results presented in sections 4 and 5.
A �nal section concludes.


2       Parameter Heterogeneity in Ordered Choice Models
The random parameter (mixed) ordered probit model allows for parameter heterogeneity
in the explanatory variables. This has several advantages over the standard ordered choice
model used for the analysis of categorical dependent variables (McKelvey and Zavoina,
1975). First, as pointed out by Greene and Hensher (2010b), the �xed parameter version of
the ordered choice model (and in fact limited dependent variable models) cannot describe
the unobserved heterogeneity likely to be present across agents. For example, if individuals
have differing preferences vis a vis a particular labor market state, individual marginal
utilities may deviate from the mean of the coefficient generated by standard models. The
variance of these utilities is potentially as informative as their average, and the objective of
the estimation is to estimate both. For instance, if there is no heterogeneity (and conditional
mean satisfaction premia are signi�cantly different from each other), it is possible to
(unambiguously) rank different employment sectors in terms of their relative desirability.
However, if there is heterogeneity, then even a substantial discount in mean satisfaction
premia may conceal that for many workers, the on-average inferior sector may be, in fact,
preferred. The ranking of sectors then becomes much more difficult. As another example,
models predicated on the notion that the informal sector has its own internal dualism
postulate substantial heterogeneity in self-reported job satisfaction within the informal sector.


Second, the panel dimension of our data also allows us to control for individual-speci�c
random effects that may be present in the data and spuriously inflate the estimated variance
in satisfaction with different job categories. We estimate a model of generalized ordered
choice that addresses both issues.7


The random coefficients (associated with job status) are assumed independently normally
distributed βi ∼ f (β , Ω) and
    7
    Further extensions of ordered choice models with observed and unobserved heterogeneity are presented
in Greene and Hensher (2010b) and Greene and Hensher (2010a)
                                                   6
                                             βi = β + W υi                                              (1)

where W is a diagonal matrix whose elements are standard deviations8 , and υi ∼ N (0, 1).


As Hensher and Greene (2003) note, the choice of the underlying parameter distribution
must be made a priori. Though we use the normal distribution here, other ones such
as the log normal, uniform, and triangular distribution have also been employed (see
Hensher and Greene, 2003; Train, 2003). Though the log normal is mathematically more
convenient and underlies the more common mixed logit, we use the normal for two reasons.
First, it permits estimation of coefficients of both signs which is essential for our analysis
of the relative desireability of different sectors (Revelt and Train, 1998).                   Second, as
Greene and Hensher (2010a) notes the imposition of a mixture of a normal distribution on
the random effect and then a log normal distribution on the parameters seems somewhat
awkward. Turning to the other options, we have no reason to believe that, for instance,
preferences, for example, are uniformly distributed.              Similarly, although the triangular
distribution is commonly used when the interest is the willingness to pay between different
attributes, this is not our focus and its functional form would impose unrealistic assumptions
about the tails of the distribution.


We further assume that the decision maker’s happiness follows an underlying random utility
or latent regression model:



                    ∗     ′
                   yit = βi xit + δ ′ zit + µi + ǫit        i = 1, ..., N t = 1, ..., Ti                (2)

                        = (β + W υi )′ xit + δ ′ zit + µi + ǫit


where xit is a vector of covariates related to job status, zit is a vector of covariates
related to other controls whose distribution is not of interest; βi and δ both contain
unknown marginal utilities. The �rst captures the heterogeneity of the population with

   8
     One could also specify the coefficients to be dependently distributed. In this case, the coefficient vector
is expressed βi = β + Lυi where L is a lower-triangular Cholesky factor of Ω, such that LL′ = Ω and where
β and L are estimated.
                                                     7
respect to satisfaction with each job type and µi represents an individual-speci�c random
effect. Note that this speci�cations nests both a standard random effects ordered probit
model, in which only the constant term is allowed to vary, and a standard ordered probit,
in which all random parameters are constrained to be zero (see Greene and Hensher (2010b)).


As is standard in the ordered choice literature, it is assumed that:


                                                                    ∗
                       yit = jT         if and only if     κj −1 < yit ≤ κj       j = 1, ..., J

                                                                             ∗
where the J outcomes are obtained by dividing the real line, represented by yit into J intervals,
using J + 1 constant but unknown threshold parameters κ0 , ..., κJ . In order to ensure well-
de�ned intervals, we need to assume ascending thresholds such that κ0 <, ..., < κJ . We code
the intervals from 1 to J and account for the ordinal nature of our data since higher values
    ∗
of yit yield higher outcomes of yit . The probability of the observed outcome sequence J for
the person i is therefore:


                                  Ti
Pr[yit = jT |xit , zit , υi ] =         F (κJ − (β + W υi )′ xit − δ ′ zit ) − F (κJ −1 − (β + W υi )′ xit − δ ′ zit )
                                  t=1
                                                                                                                  (3)
where xit contains a column of ones; F (·) is the distribution of the error terms in which
F (·) = Φ(·) if the error terms follow a standard normal distribution.9 Equation 3 contains
the unobserved random terms Ï…i which must be integrated out for estimation:


                         Pr[yit = j |xi , zi ] =        P r[yit = jT |xit , zit , Ï…i ]f (Ï…i )dÏ…i                  (4)

The model is estimated by simulated maximum likelihood (See Train, 2003, for a thorough
explanation of the estimation of discrete choice models by simulation). The simulated log
likelihood function is given by:




   9
       We need to impose the normalization κ0 = 0 in order to estimate the constant parameter.
                                                            8
                                    N               R   Ti
                                               1
          SLL(κ, β , W , δ ) =           log                  (F (κJ − (β + W υir )′ xit − δ ′ zit ) −
                                   i=1
                                               R   r =1 t=1
                                   −F (κJ −1 − (β + W υir )′ xit − δ ′ zit ))                            (5)


where Ï…ir is the rth random draw for each individual. We use Halton draws because the
literature suggests that these are superior to standard draws in this context.10 Clearly,
eliminating the stochastic component of β yields a standard ordered probit with random
effects. Likewise, if T = 1, then the simulated log likelihood collapses to a simulated
estimation of the pooled sample or cross section.


Though we also present �xed effects estimates following Ferrer-i Carbonell and Frijters
(2004) for reference (See Annex A), in general this estimator faces two problems in the
present as in other non linear contexts (See Greene and Hensher (2010a) for a more
exhaustive literature review). First, maximum likelihood estimation generates often severe
bias when T is small (the incidental parameter problem). In the binary logit model, and with
T = 2, Abrevaya (1997) has shown analytically that the full MLE estimate converges to 2β .
The Monte Carlo results in Greene (2004) suggest that biases comparable to those in binary
choice models also jaundice �xed effects ordered choice models. Second, no appropriate
procedure has been proposed in order to estimate a �xed effect ordered choice model with
random parameters.



3        Data and Descriptive Statistics

3.1      The Ghana Urban Household Panel Survey
The Ghana Urban Household Panel Survey (GUHPS) conducted by the Centre for the Study
of African Economies provides the basis for our empirical analysis. It collects information on
inter alia, incomes, education, labor market experience, household characteristics and sub-
jective wellbeing of labor force participants (ages 15 to 60) in the four largest urban centers

    10
     Bhat (2001)’s Monte Carlo analysis found the error measures of the estimated parameters was smaller
using 100 Halton Halton (1960) draws than 1000 random numbers in mixed logit models.
                                                        9
of Ghana: Accra (and neighboring Tema), Kumasi, Takoradi and Cape Coast. The sampling
scheme is based on a strati�ed random sample of urban households from the 2000 census in
Ghana. Thus, the data are roughly representative of urban labor markets. The data span a
7 year period from 2004 until 2010, but the 2007 wave is a recall wave in which no indica-
tors of subjective wellbeing were collected and is therefore excluded from the present analysis.


Personal Characteristics and Labor market Indicators: The survey contains the standard
measures of age, years of education, gender, whether the respondent is married and is the
head of the household, city of residence and ethnicity. We further construct height Z-scores
conditional on age and gender as a proxy for the general health of the respondent.


For wage employees the earnings variable measures real monthly wage income, while for
the self-employed the real earnings measure is based on enterprises pro�ts. Thus, earnings
for self-employed workers reflect both the returns to capital and the returns to labor. In
addition to these earnings measures, and hours worked, the survey includes three questions
that allow greater precision in the measure of conditional wages: tenure in the present job;
years in the formal sector, and whether or not the present position is an apprenticeship.


We are also able to construct a measure of household assets from questions on whether
the family owns a bike, telephone, electric stove, motorcycle, car etc. It does not include
�nancial assets.


Satisfaction Measures: Our subjective wellbeing measures are provided by the answer
to the question: “All things considered, how satis�ed are you with your current work?
(/life?/�nancial situation?)�. In all three cases, the options given to respondents were:
“1.Very Dissatis�ed, 2. Dissatis�ed, 3. Neither Satis�ed Nor Dissatis�ed, 4. Satis�ed, 5.
Very Satis�ed�. Information on subjective job satisfaction was collected in all waves (except
the recall wave), while information on life- and �nancial satisfaction was collected in four
waves.


Dimensions of Job Satisfaction: The survey also contains variables that help us shed light on

                                              10
the drivers of differences in job satisfaction for select subsamples. To start with, for a subset
of respondents we have information on job characteristics. Second, individuals were asked
to rate how much control they considered they had over their lives, varying from “1. No
control�, “2 some control�, “3 a great deal of control� to “4 complete control�. This variable
may capture the degree that self-employment is associated with greater empowerment, for
example by providing more independence.


Third, the survey also generated an experimentally solicited measure of risk aversion through
a series of lottery games played with a subset of the sample(see Falco et al. (2010)). A
representative subset of 288 respondents from the UPS was invited to participate to a
number of workshops where they were presented with a series of 21 lottery games. In each
game participants were asked to choose between two alternative lotteries. Each lottery took
the shape of an opaque bag containing a number of coloured marbles. To each colour was
attached a money prize and each respondent was asked to choose from which of the two bags
he would prefer to pick a random marble. At the beginning of the task each participant was
told that, once they had made their 21 choices, one would be picked at random and acted
out, i.e., they could pick a marble from the bag they chose in that case and would be paid
the corresponding winnings. The winnings of the gamble ranged from 10,000 to 110,000
CEDIs, with average winnings calibrated just above 30,000 CEDIs. This sum was estimated
to be larger than the average daily earnings of a worker in the sample and therefore the
stakes were high enough to elicit participants’ true risk-preferences. From this, an estimate
of each of the participants’ individual Coefficients of Relative Risk Aversion was retrieved.11



3.2     Employment States
For the purpose of this analysis, we distinguish between �ve paid employment states among
workers who respond to the standard question about whether they have done any work for
pay, pro�t or gain for the last seven days, even for one hour.12



  11
      For further details on how this measure was constructed, the sampling design and the de�nition of other
key variables (see CSAE, 2008; Falco, 2010; Falco et al., 2010).
   12
      We do not have a variable for whether a worker is covered by the sparse social protection programs that
exist, hence we base our categorizations of formality on �rm size.
                                                     11
Self-Employed, with Employees and Self-Employed without Employees: We divide indi-
viduals who declare themselves to be self-employed into two categories to allow greater
discrimination between those who may be more “entrepreneurial� (and hence able to hire
employees) and those who are perhaps in a disguised unemployment subsistence mode until
their next job. The division is crude, but it is preferred to a simple aggregate of the two, as
it captures an important dimension of heterogeneity. Typically, �rms in the �rst category
employ one other person, though the average number of workers employed is three. Neither
group is obligated to contribute to national social protection programs, and they are not
automatically covered. Hence, we consider them informal.


Informal Salaried and Formal Salaried Workers: We de�ne informal salaried workers to
be those working for wages in �rms with up to 5 workers. Conversely, we de�ne formal
employees as wage employees in large �rms, i.e. �rms with more than 5 workers.


Civil/Public Sector Workers: We treat workers in the public sector as a separate “formal�
category for two reasons. First, wage setting and non-monetary compensation may differ
from that in the private sector. Second, a popular perception is that public sector jobs are
the most desirable jobs. We test whether this is true.


Unpaid, Out of the Labor Force (OLF) and Unemployed: This category comprises workers
who are not working for pay.      It also includes those out of the labor force and those
unemployed. Distinguishing between these two categories is difficult given changes in survey
design over time.



3.3    Descriptive Statistics
Descriptive statistics on the different types of workers are presented in Table 1. Though
less commonly presented, summary statistics on the dynamic (transition) patterns are also
informative and we present these in the form of a transition matrix in Table 2. Students,
unpaid workers and those younger than 15 or older than 65 years of age are excluded from
our sample. Several �ndings merit note:


                                              12
First, Table 1 con�rms the stylized fact for Africa that informal jobs account for the bulk
of urban employment. In our sample, formal salaried employment accounts for only 16%
of all paid employment and civil employment accounts for a slight 8%. Self-employment
accounts for the majority (60%) of employment and thus merits the close study we bring to
it. Amongst the self-employed, approximately one out of �ve hire at least one worker. The
overwhelming majority work by themselves. The informal salaried constitute a relatively
small fraction of employment at 16%. The informal sector as a whole thus accounts for more
than three-quarters of all paid employment in our sample.


Demographically, women are much more likely to be self-employed, whereas formal and civil
employment are disproportionately male and informal salaried work is slightly more male.
Average years of education increases moving from self-employed without employees, to
self-employed with employees, to informal salaried, to formal private and civil employment.
Average age and marital status suggest important differences in where across the life cycle
workers are found, with the informal salaried being younger and seemingly less-established,
on average, at the beginning of their professional life cycle, while the self-employed are closer
to the end.13


Unconditional earnings are highest for those employed in the civil sector.                     Somewhat
surprisingly, self-employment with employees is next followed by formal salaried and then
self-employment with no employees. Informal salaried employment holds up the bottom of
the ranking. These would clearly be more informative were they conditional, although even
then, we would not be compensating for unobserved job characteristics such as degree of
independence (autonomy), risk, workplace conditions, future earnings prospects, and the like.


Table 2 presents transition matrices for movements across sectors over one year. These

  13
      The informal salaried are youngest at an average age of 31 and the formal salaried are two older at
33. The self-employed without employees are 37 years old on average; the self-employed with employees are
approximately two years older on average with a mean age of 39, which is just above the average age of the
civil employed which is 38. Only 31% of informal salaried workers are married compared to 39% for the
formal salaried, 57% for self without employees and 68% for self-employed with employees. The highest rate
of apprenticeship is found among the informal salaried. Small numbers of apprenticeships are also seen among
the self-employed (this is possible if the primary job is self-employment but the workers is also engaged in
an apprenticeship elsewhere) These patterns are remarkably similar to those found for Argentina, Brazil and
Mexico (Bosch and Maloney, 2010)
                                                    13
summarize the data underlying the panel estimators employed later and offer further insight
into the nature of the sectors. First, as has been noted in other developing countries, there
is a relatively high degree of mobility. For instance, of those declaring themselves formal
salaried at the beginning of the period, only 63% were still in that sector at the end of
the year. Second, to the degree that turnover is a measure of attachment to a sector (as
opposed to rates of �ring), formal salaried, civil employment and self-employment show
a comparatively high percentage of stayers, (low rate of turnover) at 63%, 69%, and 66%
respectively. Informal salaried employment, on the other hand, has the highest rate of
turnover; the proportion of stayers is only 45% implying that over half of the sector leaves
to other sectors in a year. This, and the demographic patterns noted above, are consistent
with it being a sector of entry of relatively young people who are shopping around for their
next (superior) employment. Further insights can be gleaned from studying the patterns
of mobilility among sectors although, as with the conditional wage comparisons discussed
earlier, they cannot offer robust conclusions about the relative desireability of the various
sectors.14


Hence, we turn to an examination of the job satisfaction data. Returning to the uncon-
ditional tabulations in Table 1, several notable facts emerge. To start with, relative to
other regions, Ghanaian workers are somewhat less satis�ed with their jobs. Using the
more homogeneous group of formal salaried workers as the comparator, Ghana yields a
value of 3.33. To put this into perspective, consider that a rudimentary rescaling of the
seven point scale used by Benz and Frey (2008b) into a comparable �ve point scale yields

  14
     For instance, consistent with the age patterns discussed above, informal salaried workers have relatively
high rates of transition to formal employment (15%) destination sector, and to self-employment (22%). Given
the diminutive size of the formal sector compared to self-employment (13% of the labor force compared to
43% with roughly equal levels of turnover), this does suggest the informal salaried do have a relative high
predisposition towards formal employment. The overall rapid turnover and propensity to move into formal
work may suggest a classic queuing story with informal salaried employment being a less desirable holding
pattern for formal employment. That said, the high level of reverse flows from formality into informality
suggests that the joint market may be behaving more competitively, and utility at the margin potentially
equated for many workers. By contrast, the self-employed show relatively little mobility into formal salaried
work or civil work and vice versa. Two hypotheses suggest themselves in this case. First, that there is a
high degree of segmentation between the self-employed and formal salaried markets, perhaps serving as a
retirement ground for older dismissed formal sector workers. Alternatively, it may be that self-employment
is the long run destination of workers once they have accumulated sufficient human and �nancial capital.
In the �rst case, the sector is likely to show lower levels of welfare. In the second they could be higher.
See (Bosch and Maloney, 2010) for a discussion of inference based on transition matrices and comparable
exercises for Latin America.
                                                     14
3.72 for advanced European countries, 3.58 for the former Eastern European countries, and
3.80 for the US. These differentials in average wellbeing across sectors may partly reflect
the long-established �nding that life satisfaction increases more generally with income
(see Clark and Senik (2010b) for a recent review). However, other factors are likely to
be important as well, as is suggested by the greater average satisfaction of workers in
Bangladesh (3.69, $US 700) which tend to be poorer.15


More striking are the relative rankings within Ghana. Civil sector employees and self-
employed individuals who employ others report the highest levels of job satisfaction. The
difference between their job satisfaction levels is not statistically signi�cant, even though
civil sector workers earn signi�cantly more than the self-employed who employ others.
Conversely, while the average earnings of wage employees in large �rms are not signi�cantly
lower than those for self-employed workers who hire other workers, their self-reported job
satisfaction is signi�cantly lower. The workers with the lowest levels of earnings and job
satisfaction are wage workers in small enterprises.        The standard deviations of the job
satisfaction measure are large pointing towards substantial heterogeneity in satisfaction
within sectors. Life and �nancial satisfaction exhibit similar patterns, though it is worthwhile
noting that the self-employed who employ other have marginally higher levels of average life
and �nancial satisfaction than public sector employees, and that the self-employed without
employees are on average more satis�ed with their �nancial situation than the formal salaried.


Clearly, it is desirable to both more rigorously model the discrete nature of the subjective
responses and to condition on individual characterstics. In the next section, we begin with
the commmon �xed parameter ordered probit to provide baseline results. We then employ
the random parameter ordered probit estimations to examine the distribution of preferences
around the mean tendencies.




  15
   A rudimentary rescaling of �ndings from Blanchflower (2000) gives an average value of 3.88 for the
OECD.
                                                15
4        Estimation

4.1      Fixed Parameter Ordered Probit
Table 3 presents our baseline estimates of job satisfaction using conventional �xed parameter
ordered probit models. In column 1, only dummies for job type are included. The omitted
category is wage-employment in large �rms.            The models have low predictive power,
explaining between 2 and 5% of the variation in self-reported satisfaction. The limited
explanatory power of such models is typical in the literature on job satisfaction and attests
to the high heterogeneity in job satisfaction across different types of jobs.


The results are broadly consistent with the statistics presented in Table 1: workers in
informal wage employment have the lowest job satisfaction on average, and self-employed
sole proprietors are also less satis�ed than wage employees in large �rms, although not
signi�cantly so. Civil sector employees are not signi�cantly more satis�ed with their jobs
than workers in the formal private sector.         Surprisingly, self-employed individuals with
employees report the highest levels of job satisfaction.16


The following columns sequentially include three blocks of additional covariates. Column
2 introduces the standard components of labor compensation; earnings, hours worked and
hours worked squared. Column 3 introduces tenure and apprenticeship variables to further
condition remuneration, as well as the household assets variable which serves as a crude
proxy for household wealth and may help mitigate potential measurement error in our
earnings variable.    The interpretation becomes then “taking into account differences in
earnings, how do the other characteristics of these jobs affect percieved wellbeing.� Column
4 includes only personal characteristics, but does not condition on compensation or job
characteristics. Finally, column 5 combines all covariates.


Conditioning on labor compensation (column 2) generates a positive and signi�cant sign
on earnings, and eliminates the negative coefficient on informal salaried employment. This


    16
    Controlling for location and year renders the dummy on working in the civil sector insigni�cant,
suggesting that location matters
                                                16
suggests that lower earnings is the primary cause of dissatisfaction with this sector. However,
being self-employed with employees remains signi�cantly positively associated with job
satisfaction even after we control for the fact that such workers tend to have higher earnings.
Note that the average utility premium associated with being a small �rm owner who employs
other is large, especially when compared to the coefficient associated with earnings; in
monetary terms, the satisfaction premium is roughly equivalent to a doubling of earnings.
These results are robust to the third block of variables (column 3), notably tenure,17
whether or not the individual is currently an apprentice, and household assets. Household
wealth is positively correlated with job satisfaction and diminishes the impact of earnings
only slightly. Being an apprentice enters positively but is not remotely statistically signi�cant.


The results are also robust to including controls for individual characteristics, such as
gender, age, education, height, marital status, being the head of the household and ethnicity.
Column 4 includes these controls alone. The resulting coefficients on the different job type
dummies can be interpreted as satisfaction differentials net of individual characteristics.
As such, this speci�cation provides a test for the segmentation hypothesis. Though the
results do suggest that being better educated, older,18 healthier (a higher height Z-score)
and married is associated with signi�cantly greater satisfaction, and being a household
head with signi�cantly lower satisfaction, the results of focal interest do not change
appreciably from the base speci�cation in column (1): Being self-employed with employees
is associated with a satisfaction premium, being informal salaried with a satisfaction discount.


Speci�cation 5 includes all explanatory variables; the coefficients on the job dummies now
reflect satisfaction differentials net of individual characteristics, earnings, job characteristics
and household assets. The satisfaction premium associated with being self-employed with
employees remains signi�cantly positive in all speci�cations, some interesting differences
with speci�cation 4 emerge; the education effect disappears suggesting that its channel
of influence is through earnings. By contrast, the gender dummy which was insigni�cant

  17
      We also experimented with speci�cations including the square of tenure, but these did not enhance the
explanatory power of the model. Results are omitted to conserve space, but available upon request from the
authors.
   18
      The quadratic term quickly swamps the free standing term leaving the minimum value at .4 years of age
and the joint impact on job satisfaction positive after age 1.
                                                    17
when earnings were not conditioned on, is signi�cantly positive now that it is controlled
for; women are happier with their jobs, ceteris paribus, perhaps because they have lower
earnings expectations (see e.g. (Clark, 1997)).


Comparing the different speci�cations, the conditioning variables seem important to the
relative ranking of sectors only for informal salaried work, with the other signi�cant
premium on self-employment with employees remaining affected relatively little.           The
discount on informal salaried work suggests that differences in job satisfaction across sectors
are predominantly driven by differences in earnings, rather than differences in workers’
observable characteristics, since the sector premia are much more sensitive to inclusion of
the latter than the former: workers in this sector appear to be earning, on average, less
than is the norm conditional on human capital. This would be consistent with the �ndings
from the summary statistics that suggest that these are often individuals just entering the
workforce who have yet to �nd a good match.




4.2    Allowing for Parameter Heterogeneity: Random Parameters
       (Mixed) Ordered Probit Estimation
The coeffients on job sectors in the previous results reflect the central tendency-the average
premium or discount enjoyed by individuals within a sector. However, they obscure the
potential latent heterogeneity across individuals within particular sectors.      The Mixed
Ordered Probit expicitly allows for such heterogeneity by estimating the variance of the
parameters around the central tendency. Table 4 re-estimates the two most informative
models from the previous exercise, generating the implicit distribution of satisfaction premia
associated with different jobs. The �rst three columns condition on individual characteristics
but not earnings thereby allowing the preference parameter to include remuneration as part
of conditional wellbeing. The next three columns condition on earnings, and the sectoral
parameters thus represent preferences over non-remunerative characteristics of the type of
employment. In each group of three we report �rst, a pooled speci�cation (Columns 1,4)
that allows for random parameters (labeled “RPOP� model in the table). Subsequently,
(in Columns 2,5) we exploit the panel dimension and allow for individual-speci�c random

                                             18
effects (labeled “REOP�). Finally, (Columns 3 and 6) we combine the two yielding the
random parameter random effect mixed model (labeled’ “RPREOP�). All speci�cations
were estimated with 100 Halton draws. Using more draws did not lead to signi�cant changes
in the estimated coefficients.


The previous results are broadly robust to allowing for parameter heterogeneity. In column
1, the mean coefficient associated with being self-employed and employing at least one other
worker remains signi�cantly positive and of similar magnitude. Informal salaried work is
again associated with an average utility discount, albeit only signi�cant at the 10 percent
level. As before the remaining job coefficients are not signi�cant. The signi�cant coefficients
on all the standard deviations of the job dummies, however, reveal that focusing just on
the central tendency alone veils useful information. For all job dummies we strongly reject
the null hypothesis that there is no heterogeneity within the sector. Self-employed with
employees has the greatest variance at .76 followed by civil employment at .75, informal
salaried at .51 and self-employment with no employees at .46.            Self-employment with
employees offers higher highs, but also potentially lower lows than other sectors. Somewhat
counterintuitively, while we might expect this to be more the case in self-employment than in
salaried work, this does not appear to be the case since the satisfaction premium associated
with civil sector salaried employment has an almost equivalent variance.


Perhaps the most likely potential explanation for the striking heterogeneity in satisfaction
is variability in earnings within a sector. However, conditioning on compensation (column
4) hardly affects the estimated standard deviations of the job satisfaction premia associated
with different sectors. While it eliminates any signi�cant average informal salaried discount
and reduces the coefficient on self-employment with employees somewhat, the differences in
parameter variance are not driven by differential earnings dispersion across sectors.


Another possibility is that this heterogeneity is driven by individuals’ latent personality
traits affecting their self-reported satisfaction with all sectors. To allow for this possibility,
we estimate models that control for individual speci�c random effects. Columns 2 and
5 �rst present speci�cations that exploit the panel dimension of the data and allow for

                                               19
individual-speci�c random effects but not for parameter heterogeneity. The variance of the
random effect is signi�cant at the 1% level in both cases. The estimated mean satisfaction
differentials are very similar to those obtained using a �xed parameter probit; the average
premium associated with being self-employed and employing others is consistently postive
and statistically signi�cant, while the premium associated with being informal salaried is
negative and statistically signi�cant at the 10% level, unless remuneration is conditioned on.19


Columns 3 and 6 then combine both random effects and random parameters. Mean sector
satisfaction premia are hardly affected. By contrast, the estimated standard deviations
associated with the job dummies all fall substantially compared to speci�cations that only
allow for random parameters, though they remain statistically signi�cant. Again, these
reductions may arise because the estimated random effect σ is capturing heterogeneity
that is common across all sectors, for instance, a general positive or negative attitude
toward whatever situation an individual �nds him/herself in. What remains in column 3
and column 6 is the heterogeneity that is particular to each individual sector, the latter,
again, conditioning on renumeration. But, as found previously, controlling for compensation
introduces only modest changes; being self-employed and employing others continues to be
associated with a highly signi�cant satisfaction premium, and again, the informal salaried
now suffers no discount. As with the RPOP, conditioning on compensation surprisingly
diminishes the estimated variance of job parameters only marginally, if at all. Little of the
variation in satisfaction appears to arise from different earnings outcomes, hours worked, or
whether the job was an apprenticeship.


To help facilitate interpretation of these results, Figure 1 plots the kernel estimates of
the distributions of each sector parameter. Centering the distribution on the estimated
conditional mean sectoral premium, Table 5 then calculates the share of the sector for whom
satisfaction relative to the formal sector is greater than zero.20 The results attest to the

  19
      The values of σ suggest that roughly 7 percent of the overall variance is due to unobserved individual
effects in the case of RE models without remuneration controls (column 2) This fraction falls with the
introduction of the remuneration variables and the random parameters
   20
      To see this recall that we can compute the proportion of the population with a positive premium as
100 × Φ β  σ , where Φ is the cumulative standard normal distribution and β and σ is the mean and the
standard deviation of the coefficient.
                                                    20
importance of studying the heterogeneity underlying the central tendency. For instance,
although our parameter estimates indicate that self-employed individuals are on average
signi�cantly happier than workers with other types of jobs, as has been found throughout
the literature, the distribution of this parameter is such that for approximately a third (32%)
of the self-employed who are employing others the satisfaction premium is negative when
compensation is not controlled for but individual-speci�c random effects are. That is, while
the majority prefer self-employment, there appear to be mis�ts of the type discussed by
Evans and Leighton (1989). Further, the large variance suggests that many of those deriving
a negative (positive) premium are very unhappy (happy) relative to the mean; taken at face
value, the compensating differential required to render those who are one-standard deviation
below the mean as satis�ed as those at the mean would be crudely equivalent to a tripling
of their earnings.


Similarly, despite no signi�cant discount or premium on average, 42% of the self-employed
who do not employ others, and 60% of civil sector employees appear to experience a positive
satisfaction premium associated with being in their respective jobs. However, the variance
on the former is now small, such that even those who might prefer to be formal are “close�
in terms of utility to those happy to be there. Although we might expect the “mis�ts� to be
found here in what the small �rm size might suggest is more subsistence-oriented work, we
actually do not �nd an especially large variance relative to what we �nd for self-employment
with employees, that would suggest this to be the case. Combining the two categories (see
Annex 1) the standard deviation is statistically signi�cant and almost double that of the
self-employed with no employees value and yielding a roughly 60% - 40% split on those more
or less happy being self-employed when we do not control for compensation. The big driver
of the variance is the relatively smaller, but much more heterogenous employer sector.


That informal employment is not necessarily considered inferior to formal sector wage
employment is also attested to by the fact that 32% of the informal salaried, despite
suffering an average utility discount, prefer to be so relative to being formal workers. Hence,
while the informal salaried may appear to be in an inferior sector on average, the proportion
thinking otherwise is quite substantial.    Further, the relatively lower variance suggests,

                                              21
again, that compared to either civil employment or self-employment with employees, the
informal salaried are not so heterogeneous. The preferences are pretty tightly concentrated
- none too thrilled, none too miserable.


Once we hold compensation constant, the conclusion that those employed in the informal
sector do not perceive these jobs as inferior (that is, they do not especially miss the
non-remunerative aspects of formality) strengthens. Two-thirds of the self-employed, both
with- and without employees would not prefer formal employment.21 Individuals in informal
salaried employment seem fairly evenly divided on the value of the non remunerative aspects
of being formal or working in larger �rms; only 47% would prefer to switch to such a job.
This may suggest that the quality of formal bene�ts is not great relative to arrangements in
the informal sector.


Taken together, we �nd that, on average, self-employment with employees enjoys a persistent
premium. The informal salaried appear less satis�ed than the formal salaried, unless we
control for incomes in which case the conditional job satisfaction differential becomes
statistically insigni�cant. Self-employment without employees and civil employment are on
average valued equally to formal salaried work. However, we �nd very strongly signi�cant
evidence for heterogeneity within each sector that undermine statements about the relative
desireability of each sector based on mean differentials alone.                In every sector we �nd
substantial subpopulations that would (not) prefer formal employment. Further, there is
also substantial heterogeneity in the degree of heterogeneity across sectors. In some, for
instance self-employed with employees, individuals with radically different levels of relative
satisfaction coexist within the sector. For the informal salaried, there is far less diversity of
opinion about the merits of the sector.



5        Robustness: Fixed Effects, Alternate Measures of
         Wellbeing, and Attrition
This section examines several factors that may be driving the distribution of satisfaction

    21
     Pooling these two groups, as is done in the appendix, further strengthens this conclusion, as only 17%
of the self-employed would prefer to be formal wage workers
                                                    22
premia. We �rst examine whether or not the results are robust to controlling for individual
�xed effects.       Subsequently, we investigate whether the results are robust to different
measures of wellbeing and whether there may be tradeoffs between them. Third, we examine
the potential importance of survival bias (attrition).


5.1     Controlling for Time Invariant Unobserved Individual
        Characteristics
Random effects models are efficient and desirable in the present context so long as there
is no correlation between unobserved-individual characteristics, i.e. personality traits, such
as being optimistic or pessimistic, and the explanatory variables. Though our stochastic
parameter model assumes an underlying normal distribution, to date, there is no �xed
effects probit estimator. As an indirect test of the robustness of the overall patterns we
identi�ed in the previous section, we compare our results with those from an ordered logit
and then a �xed effects ordered logit model proposed by Ferrer-i Carbonell and Frijters
(2004). If the results are similar with the ordered logit to our ordered probit, and then
are also robust to the inclusion of �xed effects, that gives some measure of con�dence that
unobserved individual effects are not driving our results. The �xed effect estimator clearly
forces us to exclude individuals whose job satisfaction does not vary over the period over
which they are observed, which constitutes roughly a third of our sample.22 This renders
the tradeoff between bias and efficiency involved in the choice of moving to a �xed effects
model particularly acute in this context.


The results are presented in Table 6, which again replicates the model with and without
remuneration (columns 4 and 5 from Table 3. The ordered logit results while not directly
comparable numerically, nonetheless show a virtually identical pattern of relative job
desireability to those of the analogous probit. In particular, informal salaried work suffers
a discount until we control for earnings at which point it, again, becomes insigni�cant.
Columns 3 and 4 repeat the exercise with the �xed effects estimator. Clearly, all time-

   22
     To avoid this reduction in sample size, we also replicated our results using a standard Fixed-Effects model,
which does not suffer from this drawback, yet forces us to assume cardinality. Ferrer-i Carbonell and Frijters
(2004), however, show that assuming cardinality rather than ordinality has little impact on the results. The
results which are not presented to conserve space are qualitatively similar to those obtained using the �xed
effects ordered logit model.
                                                      23
invariant characteristics are removed through �xed effects.23                 Several important results
emerge. First, the premium to self-employment with employees rises substantially relative to
the logit and remains strongly signi�cant with and without remuneration included. Second,
all other dummies are not signi�cant and in particular, we cannot reject the null that there
is no satisfaction discount associated with being informal salaried or premium associted with
being employed in the civil sector. Thus, the result that being self-employed and having
employees is associated with higher levels of job satisfaction does not appear to be driven by
workers’ unobserved personal characteristics, for instance, being more optimistic than other
labor market participants.24 If anything, our results suggest that workers in all informal
jobs are less optimistic. Hence, the �xed effect logit results attest to the overall robustness
of the observed pattern of the RERPOP results.



5.2     Life and Financial Satisfaction
We rerun the speci�cations that condition on remuneration for two other measures of
self-reported wellbeing; life and �nancial satisfaction.              While allowing us to to gauge
how broadly robust the results are to using alternative proxies for wellbeing, clearly the
responses should differ somewhat and thereby shed additional light on the quality of the
individual jobs. For instance, a possible explanation for the job satisfaction premia offered
by Benz and Frey (2008a) is that while self-employed workers have higher levels of job
satisfaction, they may have lower levels of satisfaction in other domains, such as life and
�nancial satisfaction. Tradeoffs between these different dimensions of wellbeing can arise,
for instance, when self-employed individuals face more work-family conflicts as a result
of their longer working hours (Parasuraman and Simmers, 2001), or because they have
higher income fluctuations (Carrington et al. (1996). Moreover, assessing life and �nancial
satisfaction is of interest in and of itself, since life satisfaction is a good proxy for overall
wellbeing. Financial satisfaction provides a narrower measure of utility that focuses on
only one dimension of job satisfaction, but may capture a broader view of overall family
resources as well as an intertemporal view of an individual’s wellbeing that is not being

  23
      We also remove the height variable, as we expect only very small changes over time, which are most
likely predominantly driven by measurement error.
   24
      Incidentally, note that the �xed effects speci�cation mitigates bias due to systematic underreporting or
mismeasurement of earnings of the self-employed.
                                                     24
captured by current remuneration. For instance, small businesses may take a while to grow
to pro�tability, but may then yeild higher earnings than salaried employment over the long
run.


Table 7 presents the results using the RERPOP speci�cation. For the purpose of com-
parability, the sample is restricted to workers for whom information on all measures was
available. Since life and �nancial satisfaction are reported in only 4 waves (as opposed to
6 waves for job satisfaction), the resulting samples are both smaller and have a shorter
timespan. Column 1 repeats the full speci�cation from table 4 with the reduced sample.
Column 2 present the results for life satisfaction and column 3 �nancial satisfaction. The
results remain remarkably similar across measures and to the previous �ndings. This is
particularly true for the non-random controls with the exception of being married and
being a household head which are not signi�cant predictors of job or �nancial satisfaction
conditional on the other controls, but do help predict life satisfaction. By contrast, men have
signi�cantly lower life and �nancial satisfaction than women, but similar job satisfaction,
ceteris paribus.      Education appears uncorrelated with job satisfaction (conditional on
income) but is associated with enhanced life and �nancial satisfaction.                     Earnings and
household assets appear to have virtually the same impact on all three measures.


Looking at the sectoral dummies, we �nd that self-employment with employees has an
even higher positive premium in life and �nancial satisfaction. Further, even those without
employees now enjoy a large and sign�cant �nancial satisfaction premium relative to formal
employment. One possible way of reconciling both �ndings with those previous is to see
the remuneration variables for both sets of �rms as imperfectly capturing future projected
earnings.25


Partly as a result of using a reduced sample, the estimated patterns of heterogeneity change

  25
      In particular, the �rms with no employees may be small �rms that are still growing, and require long
work days. Hence, although the intertemporal �nancial panorama may appear superior, present job and life
satisfaction may not reflect this. Informal salaried work suffers no utility discount with any variable. This
is perhaps surprising since we may imagine that, while health bene�ts, for example, may not directly affect
work satisfaction, they affect the whole family and hence life satisfaction more generally. That we do not �nd
this may suggest that the quality of the formal bene�ts is not great relative to arrangements in the informal
sector.
                                                     25
somewhat. Informal salaried workers now show no signi�cant standard deviation in any
speci�cation. By contrast, civil employment shows a modest rise in heterogeneity in job,
life and �nancial satisfaction relative to the previous exercise. The standard deviation for
self-employment with employees is also higherthan was the case in the previous exercise both
in the job and �nancial satisfaction speci�cations, but not in the life satisfaction regression.
This appears to be due to the fact that the random effect is only signi�cant in the life
satisfaction case and we know from table 4 that this reduces, by roughly the difference
observed here, the estimated variance. The same phenomenon is seen for the self-employed
with no employees which lose the signi�cance of the standard deviation in the case of life
satisfaction.


Despite the weaker power resulting from the smaller sample size, the results with the life
and �nancial satisfaction variables broadly con�rm the �ndings with the larger sample
using just job satisfaction. Self-employment of both types appears more favorable than
before.      Overall, the self-employed and informal salaried are certainly no less satis�ed
with their lives and their �nancial situation, ceteris paribus, than those in the formal
sector. Further, there does not appear to be a strong tradeoff between �nancial and life
satisfaction on the one hand, and job satisfaction on the other. Since satisfaction measures
are strongly correlated and because work satisfaction has been argued to be the most impor-
tant determinant of life satisfaction (?), these �ndings should perhaps not come as a surprise.



5.3       Survivor/Attrition Bias
Finally, we address the potential biases due to attrition from the panel.                   Each year
approximately 30% of the respondents drop out, but many of these drop-outs are temporary,
as a signi�cant share of the dropouts re-appears in later waves.26 Moreover the attrition
does not seem systemic, but instead appears driven largely by difficulties associated with
tracking people as addresses were not recorded and the initial waves of the survey did not
use GIS mapped data. Simple probit models (not presented here to conserve space) suggest
that location and survey wave are the strongest predictors of attrition; age, education and
employment status have little explanatory power. Hence we would expect little systematic
  26
       Of the 1164 individuals interviewed in 2004, 622 were still in the sample in 2010.
                                                       26
impact, especially, on the employment dummies.


     Ideally, we would explicitly model attrition. However, the estimation of the RERPOP
models is already highly computationally demanding and we therefore follow Dunne et al.
(1989) and compare the results obtained using a sample containing observations on only
those individuals whose job satisfaction was observed in all waves after they appeared in the
data for the �rst time, the so-called “long-run survivors’, with the results obtained using the
entire sample’. Comparison of the two sets of results is informative about the size of survivor
bias, as estimates obtained for the former sample suffer from maximum survivor bias. The
results for the subsample of long-run survivors are presented in Table 8 and are very similar
to those obtained for the full sample despite the expected loss in precision arising from
losing three-quarters of the sample. Most importantly, though now less signi�cant, the
coefficients associated with being self-employed and employing others is similar in RERPOP
to those found in tables 3 and 4.27 Overall, the remarkable consistency across the signi�cant
coefficients suggests that attrition effects are not large enough to drive the results.



6        Behind the Distribution of the Satisfaction Premium
The previous exercises have demonstrated that remuneration cannot explain either the
average premia associated with various sectors nor the heterogeneity observed around them.
In this section we explore the explanatory power of three variables: sector and occupation,
self-perceived control, and risk aversion. Because these variables are not available for all
individuals, for each, we present the full model (including remuneration to soak up that
influence) for the restricted sample, and then add the potential explanatory variable. Table
9 reports the results for each (set of) control(s).




    27
     The OP estimate was .27 and is now .24; the RERPOP estimate was .23 and is now .25. Though
the standard deviation is now only signi�cant for civil employment, here again the difference in estimate is
relatively small, rising from .45 in the full sample to .60 here. The other controls that enter signi�cantly are
also within the same range. Earnings were .23 before and are now .16; household assets were .07 and are now
.09.
                                                      27
6.1     Autonomy
Self-employment is often associated with greater independence.                    To assess whether the
higher levels of job satisfaction and the distribution associated with being self-employed
and employing workers stem from greater autonomy, individual’s self-perceived control
is included. Again, sample size affects the precision of the panel estimates, in this case
eliminating the sign�cance of the random effect.                 However, the “control� variable has
virtually no impact on either the mean utility premium or the standard deviation in
either self- employment category. This suggests that differences in autonomy are not the
explanation for the observed satisfaction premia or its variance.



6.2     Job Characteristics
We examine how three job characteristics may affect the observed premia. First, for wage
workers we add dummies indicating whether or not the individual in question was engaging
in manual labor and working in the manufacturing sector28 while for the self-employed we
include a dummy for being engaged in trading. Columns 3 and 4 present the results. Note
that sample selection effects again affect the precision of the estimates. Only occupation has
explanatory power,29 and its inclusion appears to raise the premium for self- employment
with employees. The other dummmies are not signi�cant and there is virtually no change
in the signi�cant standard deviation variables for either self-employment sector of civil
employment, suggesting that differences in job characteristics are not a major determinant
of the observed preference heterogeneity.



6.3     Risk Aversion
Third, we assess the possibility that the distribution of job satisfaction results from the
distribution of risk preferences, which have been demonstrated to be a predictor of sector
choice (Falco et al., 2010), using an experimentally solicited measure of risk aversion. Since
  28
      Our survey identi�es whether wage workers are manual laborers, managers, professions or clerical workers
- yet there is not sufficient variation in the latter three groups, which is why we pool them to construct a
single indicator.
   29
      For example, workers engaging in manual labor have lower signi�cantly lower levels of job satisfaction
than workers who do not. This effect loses signi�cance when �xed effects are controlled for, yet the coefficient
estimate remains positive.
                                                     28
these data are collected for only a very small number of observations, the degrees of freedom
are relatively few. Column 5 and 6 of Table 9 show all of the sectoral dummies to be
insign�cant.     However, the standard deviations are signi�cant for self-employment with
no employees and civil employment. The risk aversion variable reduces these in only the
slightest degree suggesting that, for our small sample, it is not the source of heterogeneity.30
Thus, while our sample is very small and our risk-aversion measure crude, we tentatively
conclude that differences in risk aversion are not driving differences in job satisfaction.
However, since our estimates are imprecise further research is necessary to validate this
inference.



7        Conclusion
This paper employs mixed (stochastic parameter) ordered probit estimators to characterize
the distribution of subjective wellbeing across employment sectors for Ghana, which is
broadly representative of sub-Saharan African countries. The extraordinarily detailed Ghana
Urban Household Panel Survey offers unusual insight into developing country labor markets.
The self-reported job and life satisfaction measures it collects are useful proxies for wellbeing
as they are comprehensive and rely on individuals own weighting of the importance of various
job attributes, thereby mitigating the limitations of objective proxies such as earnings. The
random parameter methodology permits quanti�cation of sectoral satisfaction premia as
well as how conditional satisfaction varies across agents within sectors, thereby allowing us
to document heterogeneity in conditional subjective wellbeing that �xed parameter models
cannot detect. They are thus a useful tool in characterizing different employment sectors,
and offer important insight into the debate over the functioning of developing country
labor markets in general and, the role of the informal sector particular. Analogous to the
debate over self-employment in the advanced countries, understanding whether (and to
what extent) the latter is, indeed, the inferior part of a segmented labor market, or offers
superior employment opportunities for those employed in it requires quanti�cation of the
    30
     In an attempt to regain some power, we also ran static speci�cations (�xed parameter-not shown to
conserve space but available upon request). To test the hypothesis that risk-averse individuals will be less
happy in self-employment where earnings are arguably more volatile, and happier in wage employment, which
is arguably less risky, we interact sector dummies with our experimentally elicited measure of risk aversion.
The resulting interactions were neither individually nor jointly signi�cant, yet the pattern of satisfaction
premia was robust to their inclusion.
                                                     29
utility workers derive from being in particular sectors on average and, we argue, also its
distribution across agents in those sectors.


Looking at �rst moments, our results suggest that, in fact, being self-employed with
employees is by far the most desirable type of employment. By contrast, workers appear
indifferent between formal salaried employment, self-employment without employees, and
civil employment. Only the informal salaried show a discount but this �nding is not robust
across estimation techniques and dissapears when conditioning on income. The non-wage
bene�ts of being formal, surprisingly, appear not to affect utility.


Moreover, the estimates of the variance of satisfaction suggest a high degree of overlap in
the distribution of job satisfaction across sectors. Despite the average premium to informal
self-employed with employees, the variance around the central tendency is very large and
roughly a third would prefer formal wage employment. Similarly, 42% of the sole proprietors
prefer to be self-employed over having a formal sector wage job even though on average there
is no signi�cant premium or discount associated with being in the sector. Conversely, about
a third of the informal salaried would prefer to be so over having a wage job in a formal
�rm, even though the mean premium associated with being informal salaried is negative. In
sum, most of the informal self-employed, and a non trivial share of the informal salaried in
Ghana show higher satisfaction than formal sector workers. While we �nd some �mis�ts�
in informal self-employment, and unsettled young people in informal salaried employment,
overall, the heterogeneity in self-reported wellbeing does not appear driven by segmentation
across sectoral lines.


These results are robust to controlling for compensation, worker �xed effects, occupational
characteristics and using alternative proxies for subjective wellbeing. However, examining
what drives the enormous heterogeneity in self-reported satisfaction we document remains a
challenge. Our exploratory foray into this important area for future research suggests that
the heterogeneity is not driven by a desire for empowerment, differences in occupational
characteristics, attitudes towards risk or tradeoffs between different dimensions of wellbeing.



                                               30
References
Abrevaya, J. (1997). The Equivalence of Two Estimators of the Fixed-Effects Logit Model.
  Economics Letters, 55(1):41–43.

Akerlof, G., Rose, A., Yellen, J., Ball, L., and Hall, R. (1988). Job Switching and Job
  Satisfaction in the US Labor Market. Brookings papers on economic activity, 1988(2):495–
  594.

Benz, M. and Frey, B. (2008a). Being Independent is a Great Thing: Subjective Evaluations
  of Self-Employment and Hierarchy. Economica-Oxford, 75(298):362–383.

Benz, M. and Frey, B. S. (2008b). The value of doing what you like: Evidence from the
  self-employed in 23 countries. Journal of Economic Behavior and Organization, 68(34):445
  – 455.

Bhat, C. (2001). Quasi-Random Maximum Simulated Likelihood Estimation of the Mixed
  Multinomial Logit Model. Transportation Research Part B: Methodological, 35(7):677–693.

Blanchflower, D. (2000). Self-employment in OECD Countries. Labour economics, 7(5):471–
  505.

Blanchflower, D. and Oswald, A. (1998). What Makes an Entrepreneur? Journal of Labor
  Economics, pages 26–60.

Boes, S. and Winkelmann, R. (2006). The Effect of Income on Positive and Negative
  Subjective Well-Being. University of Zurich, Socioeconomic Institute, Manuscript, IZA
  Discussion Paper.

 oo, D., Madrigal, L., and Pag´
B´                            es, C. (2010). Part-Time Work, Gender and Job Satisfaction:
  Evidence from a Developing Country. The Journal of Development Studies, 46(9):1543–
  1571.

Bosch, M. and Maloney, W. (2006). Gross Worker Flows in the Presence of Informal Labor
  Markets: Evidence from Mexico, 1987-2002. Policy Research Working Paper, 3833.

Bosch, M. and Maloney, W. (2010). Comparative Analysis of Labor Market Dynamics Using
  Markov Processes: An Application to Informality. Labour Economics, 17(4):621–631.
                                          31
Carrington, W. J., McCue, K., and Pierce, B. (1996).        The role of employer-employee
  interactions in labor market cycles: Evidence from the self-employed. Journal of Labor
  Economics, 14(4):571–602.

Clark, A. (2004). What Makes a Good Job? Evidence from OECD Countries. DELTA
  Working Papers.

Clark, A. and Senik, C. (2010a). Who Compares to Whom? The Anatomy of Income
  Comparisons in Europe*. The Economic Journal, 120(544):573–594.

Clark, A. E. (1997). Job satisfaction and gender: Why are women so happy at work? Labour
  Economics, 4(4):341–372.

Clark, A. E. and Senik, C. (2010b). Will gdp growth increase happiness in developing
  countries? PSE Working Papers halshs-00564985, HAL.

CSAE (2008). The Ghanna and Tanzania Urban Household Panel Surveys: 2004-2006.
  unpublished manuscript.

Cunningham, W. V. and Maloney, W. F. (2001).                Heterogeneity among mexicos
  microenterprises: An application of factor and cluster analysis. Economic Development
  and Cultural Change, 50(1):pp. 131–156.

Das, M. and Van Soest, A. (1999). A Panel Data Model for Subjective Information on
  Household Income Growth. Journal of Economic Behavior & Organization, 40(4):409–
  426.

Evans, D. S. and Leighton, L. S. (1989). Some empirical aspects of entrepreneurship. The
  American Economic Review, 79(3):pp. 519–535.

                oderbom, M., and Benhassine, N. (2009). Wage Gaps and Job Sorting in
Fafchamps, M., S¨
  African Manufacturing. Journal of African Economies, 18(5):824.

Falco, P. (2010). Risk-Aversion and Occupational Choices: Evidence from Matched Field
  Experiments and Survey Data.

Falco, P., Kerr, A., Rankin, N., Sandefur, J., and Teal, F. (2010). The Returns to Formality
  and Informality in Urban Africa.
                                            32
Ferrer-i Carbonell, A. and Frijters, P. (2004).    How Important is Methodology for the
  Estimates of the Determinants of Happiness? The Economic Journal, 114(497):641–659.

Fields, G. (2005). A guide to multisector labor market models. Social Protection Discussion
  Paper Series 0505, The World Bank.

Freeman, R. (1978). Job Satisfaction as an Economic Variable.

Frey, B. and Stutzer, A. (2002). What Can Economists Learn from Happiness Research?
  Journal of Economic literature, 40(2):402–435.

Gong, X. and Van Soest, A. (2002). Wage Differentials and Mobility in the Urban Labour
  Market: a Panel Data Analysis for Mexico. Labour Economics, 9(4):513–529.

Gong, X., Van Soest, A., and Villagomez, E. (2004). Mobility in the Urban Labor Market: a
  Panel Data Analysis for Mexico. Economic Development and Cultural Change, 53(1):1–36.

Greene, W. (2004).    The Behaviour of the Maximum Likelihood Estimator of Limited
  Dependent Variable Models in the Presence of Fixed Effects. Econometrics Journal,
  7(1):98–119.

Greene, W. and Hensher, D. (2010a). Modeling Ordered Choices: A Primer. Cambridge
  Univ Pr.

Greene, W. and Hensher, D. (2010b). Ordered Choices and Heterogeneity in Attribute
  Processing. Journal of Transport Economics and Policy (JTEP), 44(3):331–364.

Halton, J. (1960). On the Efficiency of Certain Quasi-Random Sequences of Points in
  Evaluating Multi-Dimensional Integrals. Numerische Mathematik, 2(1):84–90.

Hensher, D. and Greene, W. (2003). The Mixed Logit Model: The State of Practice.
  Transportation, 30(2):133–176.

Idson, T. (1990). Establishment Size, Job Satisfaction and the Structure of Work. Applied
  economics, 22(8):1007–1018.

ILO (2002). Women and Men in the Informal Economy: A Statistical Picture Geneva.


                                            33
Kingdon, G., Sandefur, J., and Teal, F. (2005). Patterns of Labor Demand in Sub-Saharan
  Africa. A Review Paper (Oxford: CSAE).

Launov, A. and Gunther, I. (2012). Competitive and Segmented Informal Labor Markets.
  Journal of Development Economics.

Maloney, W. (1999).     Does Informality Imply Segmentation in Urban Labor Markets?
  Evidence from Sectoral Transitions in Mexico.      The World Bank Economic Review,
  13(2):275.

McKelvey, R. and Zavoina, W. (1975). A Statistical Model for the Analysis of Ordinal Level
  Dependent Variables. Journal of Mathematical Sociology, 4(1):103–120.

Mulinge, M. and Mueller, C. (1998). Employee Job Satisfaction in Developing Countries:
  The Case of Kenya. World Development, 26(12):2181–2199.

Oswald, A. (1997).      Happiness and Economic Performance.      The Economic Journal,
  107(445):1815–1831.

   es, C., Madrigal, L., and Dept, I.-A. D. B. R. (2008). Is Informality a Good Measure of
Pag´
  Job Quality?: Evidence from Job Satisfaction Data. Inter-American Development Bank
  Washington, DC.

Parasuraman, S. and Simmers, C. A. (2001). Type of employment, work-family conflict and
  well-being: a comparative study. Journal of Economic Behavior and Organization.

Perry, G., Maloney, W. F., Arias, Omar, M. A., and Saavedra, J. (2007). Informality: Exit
  and Exclusion. World Bank.

Pratap, S. and Quintin, E. (2006). Are Labor Markets Segmented in Developing Countries?
  a Semiparametric Approach. European Economic Review, 50(7):1817–1841.

Ravallion, M. and Lokshin, M. (2002). Self-rated economic welfare in russia. European
  Economic Review, 46(8):1453–1473.

Revelt, D. and Train, K. (1998). Mixed Logit with Repeated Choices: Households’ Choices
  of Appliance Efficiency Level. Review of Economics and Statistics, 80(4):647–657.

                                           34
 oderbom, M., Teal, F., Wambugu, A., and Kahyarara, G. (2006). The Dynamics of Returns
S¨
  to Education in Kenyan and Tanzanian Manufacturing*. Oxford Bulletin of Economics
  and Statistics, 68(3):261–288.

Train, K. (2003). Discrete Choice Methods with Simulation. Cambridge Univ Pr.

Winkelmann, L. and Winkelmann, R. (1998).      Why are the Unemployed so Unhappy?
  Evidence from Panel Data. Economica, 65(257):1–15.

Wright, T. and Staw, B. (1999). Affect and Favorable Work Outcomes: Two Longitudinal
  Tests of the Happy–Productive Worker Thesis.     Journal of Organizational Behavior,
  20(1):1–23.




                                          35
A      Fixed Effect Model in Ordered Choice Models
Several proposals for bias reduction estimators in the binary choice model have been made
(Winkelmann and Winkelmann, 1998; Das and Van Soest, 1999; Boes and Winkelmann,
2006). In this paper, we use the approach proposed by Ferrer-i Carbonell and Frijters (2004)
which is a modi�ed version of the model proposed by Winkelmann and Winkelmann (1998).
In this model:



                      ∗
                     yit = αi + β ′ xit + δ ′ zit + ǫit                                                              (6)
                                                     ∗
                     yit = j         if   κj −1,i < yit ≤ κj,i             j = 1, ..., J, t = 1, ..., Ti


The ordered logit form is assumed. Therefore, the probability for individual i is:


                                 Pr[νit = 1|xi , zi ] = Λ(λi + β ′ xit + δ ′ zit )                                   (7)

where νit is de�ned with respect to an individual speci�c ji∗ :


                                  νit = 1      if    yit > ji∗          and 0 otherwise

The resulting contribution to the likelihood for individual i is:


                                                                                                           
                                                                                           Ti
            Pr yi1 > ji∗ , ..., yi,Ti > ji∗ 1|xi1 , ..., xi,Ti , zi1 , ..., zi,Ti ,            νit = ci  =
                                                                                          t=1

                                                                   Ti
                                                      exp          t=1   νit (β ′ xit + δ ′ zit )
                                     =                                                                               (8)
                                                                                 Ti
                                            (ν1 ,...,νTi )∈S (ji
                                                               ∗ ,c )
                                                                   i
                                                                        exp      t=1   νi (β ′ xit   +   δ ′ zit )

where ci =       t   νit is the number of times yit is greater than the chosen threshold. The
threshold ji∗ is chosen so that ci is not equal to 0 or Ti . S (ji∗ , ci ) is the set of all possible
vectors, (ν1 , ν2 , ..., νTi ), whose elements are all zero or one and sum to ci ; that is, the set of
vectors corresponding to sets of outcomes yit such that ci of them are greater than ji∗ .



                                                              36
B   Tables




             37
                                                              Table 1: Summary Statistics: Ghana Urban Household Panel
                                                              Survey (GUHPS)

                                                    Self-employed, employees            Self-employed, no Employees              Informal Salaried          Formal Salaried              Civil
                                                    Mean           SD                   Mean             SD                      Mean       SD              Mean       SD           Mean       SD
      Job Satisfaction                               3.49          1.03                  3.27            1.00                     3.18      1.01             3.33      0.94          3.53     1.04
      Life Satisfaction                              3.57          0.85                  3.23            0.89                     3.13      0.90             3.30      0.90          3.49     0.92
      Financial Satisfaction                         2.89          1.08                  2.67            0.97                     2.46      0.88             2.59      0.97          2.82     1.08
      Male                                           0.40          0.49                  0.26            0.44                     0.55      0.50             0.75      0.43          0.63     0.48
      Education                                      8.35          3.80                  7.52            4.04                     8.64      3.58            10.20      3.16         11.43 3.16
      Age                                           39.35         10.16                 36.94           10.15                    30.76     10.04            33.22     10.25         38.33 11.73
      Height                                         0.07          0.92                 -0.01            1.08                     0.01      0.85            -0.01      0.90          0.23     0.93
      married                                        0.68          0.47                  0.57            0.49                     0.31      0.46             0.39      0.49          0.56     0.50
      Household head                                 0.55          0.50                  0.45            0.50                     0.43      0.50             0.52      0.50          0.62     0.49
      Apprentice                                     0.01          0.08                  0.00            0.06                     0.10      0.29             0.06      0.24          0.01     0.11
      Tenure (log)                                   1.85          1.01                  1.89            1.01                     1.12      0.74             1.44      0.84          1.49     1.07
      Hours (log)                                    3.86          0.53                  3.79            0.52                     3.83      0.48             3.78      0.40          3.70     0.38
      Earnings (log)                                 3.88          1.27                  3.26            1.05                     3.14      0.83             3.77      0.85          4.06     0.80
      Household Assets (log)                         5.38          1.29                  4.80            1.42                     4.74      1.43             5.12      1.47          5.30     1.24
      Control                                        2.32          0.81                  2.26            0.85                     2.15      0.85             2.41      0.78          2.49     0.75
      Bad mood                                       0.05          0.22                  0.15            0.36                     0.13      0.33             0.09      0.29          0.10     0.30
38




      Good mood                                      0.12          0.32                  0.06            0.24                     0.05      0.21             0.08      0.26          0.10     0.31
      Wage and manufacturing                         0.00          0.00                  0.00            0.00                     0.18      0.38             0.39      0.49         0.06      0.25
      Self and trade                                 0.39          0.49                  0.62            0.49                     0.00      0.00             0.00      0.00          0.00     0.00
      Wage and not a manutal laborer                 0.00          0.00                  0.00            0.00                     0.26      0.44             0.46      0.50          0.68     0.47
      Percentage                                              13.11                                46.80                               16.13                      15.73                  8.14
      Observations                                             399                                  1520                                523                        510                    264
     Note: The GUHPS covers labor force participants ages 15 to 60 in the four largest urban centers of Ghana: Accra (and neighboring Tema), Kumasi, Takoradi and Cape Coast. The
     sampling scheme is based on a strati�ed random sample of urban households from the 2000 census in Ghana and is thus broadly representative of urban labor markets. The data span
     a 7 year period from 2004 until 2010, but the 2007 wave is a recall wave in which no indicators of subjective wellbeing were collected and is therefore excluded from the present analysis.
     The �ve labor market states are Self-no employees (those declaring themselves self-employed, but with no employees); Self-with employees (those with employees); Informal salaried
     (employees in �rms of up to 5 workers); Formal salaried (employees in �rms over 5 workers); and Civil (civil or public sector workers). The �nal state combines Unpaid, out of the labor
     force (OLF) and unemployed
                                  Table 2: One Year Transitions across Employment States

                                                                     Current State
                                 Self-employed     Self-employed    Salaried   Salaried    Civil    OLF, unemployed      Total
 Initial State                     employess       no employees     Formal     Informal                or unpaid
 Self-employed, employees              226               130           17          18         2            45             438
                                       51.6             29.68         3.88       4.11      0.46          10.27            100
 Self-employed, no employees           205              1,076          93          34         7           219            1,634
                                      12.55             65.85         5.69       2.08      0.43           13.4            100
 Informal Salaried                      43                77          241          80        12            84             537
                                       8.01             14.34        44.88       14.9      2.23          15.64            100
 Formal Salaried                        11                28           82         356        28            61             566
                                       1.94              4.95        14.49       62.9       4.95         10.78            100
 Civil                                   1                 9            8          35       159            18             230
                                       0.43              3.91         3.48       15.22     69.13          7.83            100
 OLF, unemployed or unpaid              69               245          122         108        31           910            1,485
                                      4.65               16.5         8.22       7.27       2.09         61.28            100
 Total                                 555              1,565         563         631       239          1,337           4,890
                                      11.35               32         11.51        12.9      4.89         27.34            100

Note: Absolute numbers of workers transiting between initial sector i and terminal sector j (Rows sum to total in initial state);
and probability of transiting from i to j below (Pij ; Rows sum to 100%). The �ve labor market states are Self-no employees
(those declaring themselves self-employed, but with no employees); Self-with employees (those with employees); Informal salaried
(employees in �rms of up to 5 workers); Formal salaried (employees in �rms over 5 workers); and Civil (civil or public sector
workers). The �nal state combines Unpaid, out of the labor force (OLF) and unemployed. Distinguishing among these categories
is difficult given changes in survey design over time.




                                                              39
                                     Table 3: Job Satisfaction: Ordered Probit Estimates

                                                     (1)          (2)         (3)          (4)          (5)
                 Self-employed, employees        0.219***     0.218***    0.224***     0.255***     0.212***
                                                  (0.072)      (0.072)     (0.074)      (0.075)      (0.076)
                 Self-employed, no employees       -0.070        0.031      0.069        -0.023        0.046
                                                  (0.055)      (0.056)     (0.058)      (0.061)      (0.062)
                 Informal Salaried               -0.141**       -0.012      0.004        -0.121       -0.019
                                                  (0.066)      (0.068)     (0.068)      (0.068)      (0.069)
                 Civil                              0.134        0.059      0.050         0.094        0.040
                                                  (0.083)      (0.084)     (0.084)      (0.085)      (0.085)
                 Earnings (log)                               0.214***    0.204***                  0.221***
                                                               (0.020)     (0.020)                   (0.021)
                 Hours (log)                                   0.470*       0.416                      0.399
                                                               (0.257)     (0.258)                   (0.259)
                 Hours (log)2                                  -0.068*      -0.060                    -0.056
                                                               (0.039)     (0.039)                   (0.040)
                 Tenure (log)                                               -0.021                    -0.002
                                                                           (0.021)                   (0.022)
                 Apprentice                                                  0.121                     0.061
                                                                           (0.117)                   (0.120)
                 Household Assets (log)                                   0.093***                  0.090***
                                                                           (0.015)                   (0.015)
                 Male                                                                     0.019      -0.078*
                                                                                        (0.044)      (0.045)
                 Age                                                                    -0.024*    -0.040***
                                                                                        (0.013)      (0.013)
                 Age2 /100                                                               0.029*     0.047***
                                                                                        (0.016)      (0.016)
                 Education                                                               -0.015       -0.004
                                                                                        (0.014)      (0.014)
                 Education2 /100                                                        0.203**        0.029
                                                                                        (0.095)      (0.097)
                 Height (Z-score)                                                       0.039**        0.031
                                                                                        (0.019)      (0.019)
                 Married                                                                0.085**        0.037
                                                                                        (0.043)      (0.044)
                 Household head                                                         -0.079*       -0.047
                                                                                        (0.042)      (0.044)
                 Ethnicity Dummies                  No          No           Yes           Yes          yes
                 City Dummies                       Yes         Yes          Yes           Yes          yes
                 Year Dummies                       Yes         Yes          Yes           Yes          yes
                 LL                              -4234.079   -4171.594    -4149.513    -4221.036   -4138.924
                 Finite Sample AIC                 2.622       2.585        2.576        2.621         2.575
                 HQ IC                             2.633       2.598        2.593        2.640         2.597
                 Pseudo R2                         0.024       0.039        0.044        0.027         0.046
                 N                                 3242        3242         3242          3242         3242

Note: Estimation by Ordered Probit. The dependent variable is Job satisfaction which is an ordered variable with the following
categories: 1.Very Dissatis�ed, 2. Dissatis�ed, 3. Neither Satis�ed Nor Dissatis�ed, 4. Satis�ed, 5. Very Satis�ed. The
�ve labor market states are Self-no employees (those declaring themselves self-employed, but with no employees); Self-with
employees (those with employees); Informal salaried (employees in �rms of up to 5 workers); Formal salaried (employees in �rms
over 5 workers); and Civil (civil or public sector workers). Omitted category is Formal salaried. Information on subjective job
satisfaction was collected in all waves except the recall wave. Finite Sample AIC corresponds to AIC + 2M (M + 1)/(n − M − 1)
and HQ IC corresponds to (−2 log L + 2M log log n)/n where M is the number of parameters in the model. Thresholds nor
reported. Standard errors in parentheses. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01.




                                                             40
                            Table 4: Job Satisfaction: Random Parameter Ordered Probit Models

                                                         Without Remuneration                      With Remuneration
                                                      (1)          (2)        (3)             (4)           (5)        (6)
                                    Parameter      RPOP         REOP     RERPOP            RPOP          REOP     RERPOP
    Self-employed, employees          Mean        0.317***     0.255***   0.271***        0.265***      0.213***   0.230***
                                                   (0.077)      (0.075)    (0.078)         (0.078)       (0.075)    (0.078)
                                        SD        0.757***                0.589***        0.725***                 0.544***
                                                   (0.056)                 (0.054)         (0.056)                  (0.055)
    Self-employed, no employees       Mean          -0.016       -0.020     -0.021           0.059         0.047      0.049
                                                   (0.063)      (0.066)    (0.065)         (0.064)       (0.066)    (0.064)
                                        SD        0.463***                0.102***        0.421***                 0.117***
                                                   (0.029)                 (0.029)         (0.029)                  (0.029)
    Informal Salaried                 Mean         -0.122*      -0.120*    -0.122*          -0.012        -0.020     -0.020
                                                   (0.070)      (0.073)    (0.072)         (0.071)       (0.073)    (0.073)
                                        SD        0.508***                0.255***        0.505***                 0.271***
                                                   (0.048)                 (0.048)         (0.047)                  (0.048)
    Civil                             Mean           0.131        0.101      0.108           0.069         0.042      0.053
                                                   (0.086)      (0.084)    (0.087)         (0.087)       (0.083)    (0.087)
                                        SD        0.752***                0.414***        0.763***                 0.450***
                                                   (0.070)                 (0.070)         (0.070)                  (0.071)
    Earnings (log)                    Mean                                                0.248***      0.224***  0.227****
                                                                                           (0.021)       (0.020)    (0.021)
    Hours (log)                       Mean                                                  0.455*        0.421*     0.456*
                                                                                           (0.252)       (0.248)    (0.250)
    Hours   (log)2                    Mean                                                 -0.063*        -0.060    -0.064*
                                                                                           (0.038)       (0.038)    (0.038)
    Tenure (log)                      Mean                                                  -0.006        -0.001     -0.003
                                                                                           (0.023)       (0.023)    (0.023)
    Apprentice                        Mean                                                   0.070         0.054      0.056
                                                                                           (0.132)       (0.133)    (0.134)
    Household Assests (log)           Mean                                                0.096***      0.092***   0.090***
                                                                                           (0.016)       (0.016)    (0.016)
    Male                              Mean           0.019        0.023        0.022       -0.088*       -0.076*    -0.077*
                                                   (0.045)      (0.044)      (0.044)       (0.046)       (0.045)    (0.045)
    Age                               Mean        -0.027**      -0.023*      -0.024*     -0.044***     -0.040***  -0.041***
                                                   (0.013)      (0.013)      (0.013)       (0.014)       (0.013)    (0.013)
    Age2    /100                      Mean         0.033**       0.028*       0.030*      0.053***      0.046***   0.049***
                                                   (0.017)      (0.017)      (0.016)       (0.017)       (0.017)    (0.017)
    Education                         Mean          -0.014       -0.018       -0.014       0.0001         -0.005     -0.001
                                                   (0.015)      (0.015)      (0.015)       (0.015)       (0.015)    (0.015)
    Education2 /100                   Mean         0.209**      0.227**      0.204**         0.007         0.041     0.016
                                                   (0.097)      (0.097)      (0.097)       (0.098)       (0.099)    (0.099)
    Height (Z-score)                  Mean         0.044**      0.044**      0.045**       0.035*        0.034*      0.035*
                                                   (0.020)      (0.021)      (0.021)       (0.020)       (0.020)    (0.020)
    Married                           Mean         0.092**      0.088**      0.089**         0.039         0.037     0.040
                                                   (0.044)      (0.044)      (0.044)       (0.045)       (0.045)    (0.045)
    Household head                    Mean        -0.099**     -0.086**     -0.094**        -0.065        -0.052     -0.059
                                                   (0.043)      (0.043)      (0.043)       (0.044)       (0.043)    (0.043)
    σ                                                          0.283***     0.222***                    0.232***   0.141***
                                                                (0.019)      (0.019)                     (0.019)    (0.019)
    Ethnicity Dummies                                Yes           Yes          Yes           Yes           Yes        Yes
    City Dummies                                     Yes           Yes          Yes           Yes           Yes        Yes
    Year Dummies                                     Yes           Yes          Yes           Yes           Yes        Yes
    LL                                            -4210.059   -4214.922    -4207.306     -4127.878     -4136.079  -4129.292
    Finite Sample AIC                               2.617         2.618        2.616         2.570         2.573      2.572
    HQ IC                                           2.638         2.638        2.638         2.595         2.597      2.598
    N                                               3242          3242         3242          3242          3242       3242
    Individuals                                     3242          1434         1434          3242          1434       1434

Note: Estimation by Random Parameter Ordered Probit (RPOP), Random Effects Ordered Probit (REOP), Random Effects
Random Parameter Ordered Probit (RERPOP). The dependent variable is Job satisfaction which is an ordered variable with
the following categories: 1.Very Dissatis�ed, 2. Dissatis�ed, 3. Neither Satis�ed Nor Dissatis�ed, 4. Satis�ed, 5. Very Satis�ed.
The �ve labor market states are Self-no employees (those declaring themselves self-employed, but with no employees); Self-with
employees (those with employees); Informal salaried (employees in �rms of up to 5 workers); Formal salaried (employees in
�rms over 5 workers); and Civil (civil or public sector workers). Omitted category is Formal salaried. Information on subjective
job satisfaction was collected in all waves except the recall wave. Parameters for all job types are assumed to follow a normal
distribution. Random parameters estimates were estimated using 100 Halton draws. σ corresponds to random effect parameter
which is estimated assuming a random constant. Finite Sample AIC corresponds to AIC + 2M (M + 1)/(n − M − 1) and HQ
IC corresponds to (−2 log L + 2M log log n)/n where M is the number of parameters in the model. Thresholds nor reported.
Standard errors in parentheses. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01.




                                                              41
                               Table 5: Share of Sector that Would not Prefer Formal Employment

                                                       Model (1)   Model (3)    Model (4)    Model (6)
                       Self-employed, employees          66%         68%          64%          66%
                       Self-employed, no employees       49%         42%          56%          66%
                       Informal Salaried                 41%         32%          49%          47%
                       Civil                             57%         60%          54%          55%

Note: Computed as the proportion of the population with a positive premium as 100 × Φ(βk /σk ), where Φ is the cumulative
standard normal distribution and βk and σk is the mean and the standard deviation of the coefficient as estimated in Table 4

                          Table 6: Robustness to Inclusion of Individual Effects: Fixed Effects Logit

                                                             Ordered Logit            Fixed Effects
                                                            (1)          (2)         (3)         (4)
                         Self-employed, employees       0.466***     0.409***    0.715***     0.694**
                                                         (0.139)      (0.140)     (0.263)     (0.278)
                         Self-employed, no employees      -0.070       0.065       0.278       0.318
                                                         (0.101)      (0.105)     (0.231)     (0.246)
                         Informal Salaried              -0.242**       -0.060      0.214       0.283
                                                         (0.111)      (0.113)     (0.222)     (0.225)
                         Civil                             0.154        0.055      -0.035      -0.147
                                                         (0.165)      (0.167)     (0.328)     (0.331)
                         Earnings (log)                              0.410***                0.320***
                                                                      (0.041)                 (0.062)
                         Hours (log)                                    0.577                   1.231
                                                                      (0.506)                 (0.769)
                         Hours (log)2                                  -0.081                  -0.177
                                                                      (0.078)                 (0.117)
                         Tenure (log)                                  -0.008                   0.005
                                                                      (0.039)                 (0.078)
                         Apprentice                                     0.195                  -0.205
                                                                      (0.178)                 (0.388)
                         Household Assets (log)                      0.163***                0.194***
                                                                      (0.025)                 (0.050)
                         Male                              0.018      -0.154*
                                                         (0.079)      (0.080)
                         Age                              -0.036    -0.062***      -0.304     0.190
                                                         (0.023)      (0.023)     (0.550)    (0.572)
                         Age2 /100                         0.043      0.072**
                                                         (0.029)      (0.029)
                         Education                        -0.033       -0.011
                                                         (0.025)      (0.024)
                         Education2 /100                 0.408**        0.096
                                                         (0.178)      (0.175)
                         Height (Z-score)                0.062*         0.047
                                                         (0.034)      (0.035)
                         Married                          0.130*        0.033       0.039     -0.057
                                                         (0.078)      (0.078)     (0.174)    (0.178)
                         Household head                   -0.120       -0.067      -0.137     -0.136
                                                         (0.076)      (0.079)     (0.135)    (0.143)
                         Ethnicity Dummies                  Yes          Yes         No         No
                         City Dummies                       Yes          Yes         No         No
                         Year Dummies                       Yes          Yes         Yes        Yes
                                                           3,242        3,242       2,130      2,130
                         Pseudo R2                         0.029        0.050       0.072      0.103

Note: Estimation by Ordered Logit and Fixed Effects Logit. The dependent variable is Job satisfaction which is an ordered
variable with the following categories: 1.Very Dissatis�ed, 2. Dissatis�ed, 3. Neither Satis�ed Nor Dissatis�ed, 4. Satis�ed,
5. Very Satis�ed. The �ve labor market states are Self-no employees (those declaring themselves self-employed, but with no
employees); Self-with employees (those with employees); Informal salaried (employees in �rms of up to 5 workers); Formal
salaried (employees in �rms over 5 workers); and Civil (civil or public sector workers). Omitted category is Formal salaried.
Information on subjective job satisfaction was collected in all waves except the recall wave. Thresholds nor reported. Standard
errors in parentheses. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01.




                                                             42
                         Table 7: Robustness to Alternative Measures of Subjective Wellbeing
                         (RERPOP)

                                                        Job Satisfaction    Life Satisfaction   Financial Satisfaction
                                          Parameter             (1)                  (2)                  (3)
          Self-employed, employees          Mean             0.179*             0.282***              0.449***
                                                             (0.107)              (0.103)              (0.102)
                                              SD           0.708***             0.463***              0.661***
                                                             (0.077)              (0.075)              (0.074)
          Self-employed, no employees        Mean             -0.003               -0.050             0.237***
                                                             (0.089)              (0.085)              (0.083)
                                              SD           0.224***                 0.001             0.246***
                                                             (0.039)              (0.042)              (0.040)
          Informal Salaried                  Mean             -0.070               -0.100               0.002
                                                             (0.097)              (0.094)              (0.101)
                                              SD               0.004                0.098                0.009
                                                             (0.063)              (0.062)              (0.070)
          Civil                              Mean              0.054                0.107                0.050
                                                             (0.129)              (0.125)              (0.120)
                                              SD            0.797***             0.823***             0.812***
                                                             (0.106)              (0.108)              (0.106)
          Earnings (log)                     Mean          0.253***             0.225***              0.277***
                                                             (0.029)              (0.029)              (0.028)
          Hours (log)                        Mean            0.539*                 0.424                0.311
                                                             (0.303)              (0.310)              (0.304)
          Hours (log)2                       Mean            -0.081*               -0.068               -0.047
                                                             (0.048)              (0.049)              (0.047)
          Tenure (log)                       Mean              0.020               -0.014               -0.021
                                                             (0.032)              (0.032)              (0.032)
          Apprentice                         Mean              0.073                0.084                0.221
                                                             (0.154)              (0.170)              (0.178)
          Household Assests (log)            Mean          0.088***             0.094***              0.090***
                                                             (0.023)              (0.022)              (0.023)
          Male                               Mean             -0.042            -0.253***            -0.208***
                                                             (0.065)              (0.064)              (0.062)
          Age                                Mean           -0.048**            -0.066***             -0.046**
                                                             (0.019)              (0.018)              (0.019)
          Age2 /100                          Mean            0.055**             0.077***              0.055**
                                                             (0.024)              (0.023)              (0.024)
          Education                          Mean             -0.030               -0.019               -0.028
                                                             (0.022)              (0.023)              (0.022)
          Education2 /100                    Mean              0.215              0.297*               0.298**
                                                             (0.152)              (0.159)              (0.146)
          Height (Z-score)                   Mean              0.019                0.025              0.00001
                                                             (0.028)              (0.028)              (0.026)
          Married                            Mean              0.034             0.198***               -0.038
                                                             (0.063)              (0.064)              (0.062)
          Household head                     Mean             -0.088                0.003                0.012
                                                             (0.059)              (0.060)              (0.058)
          σ                                                    0.019             0.204***                0.029
                                                             (0.026)              (0.027)              (0.026)
          Ethnicity Dummies                                     Yes                  Yes                  Yes
          City Dummies                                          Yes                  Yes                  Yes
          Year Dummies                                          Yes                  Yes                  Yes
          LL                                               -2234.983            -2091.045            -2198.836
          Finite Sample AIC                                    2.566                2.403                2.525
          HQ IC                                                2.606                2.443                2.565
          N                                                    1771                 1771                 1771
          Individuals                                          1209                 1209                 1209

Note: Estimation by Random Effects Random Parameter Ordered Probit (RERPOP). The dependent variables are Job, Life and
Financial satisfaction which are ordered variables with the following categories: 1. Very Dissatis�ed, 2. Dissatis�ed, 3. Neither
Satis�ed nor Dissatis�ed 4. Satis�ed, 5. Very Satis�ed. Sample common across all estimates and is limited by the fact that
Life and Financial satisfaction measures were collected only over 4 waves. The �ve labor market states are Self-no employees
(those declaring themselves self-employed, but with no employees); Self-with employees (those with employees); Informal salaried
(employees in �rms of up to 5 workers); Formal salaried (employees in �rms over 5 workers); and Civil (civil or public sector
workers). Omitted category is Formal salaried. Parameters for all job types- Self with employees Self-no employees, Informal
Salaried and Civil are assumed to follow a normal distribution. Omitted category is Formal employment. Estimates using 100
Halton draws. σ corresponds to random effect parameter which is estimated assuming a random constant. Finite Sample AIC
corresponds to AIC + 2M (M + 1)/(n − M − 1) and HQ IC corresponds to (−2 log L + 2M log log n)/n where M is the number
of parameters in the model. Thresholds nor reported. Standard errors in parentheses. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01.




                                                              43
                                        Table 8: Robustness to Attrition: Job Satisfaction

                                                              Parameter         (1)           (2)
                              Self-employed, employees          Mean           0.239       0.250*
                                                                              (0.151)      (0.150)
                                                                    SD                       0.194
                                                                                           (0.136)
                              Self-employed, no employees        Mean          0.158        0.150
                                                                              (0.132)      (0.155)
                                                                    SD                       0.303
                                                                                           (0.218)
                              Informal Salaried                  Mean          0.122     0.679***
                                                                              (0.147)      (0.100)
                                                                    SD                       0.048
                                                                                           (0.061)
                              Civil                              Mean          0.183     0.284***
                                                                              (0.203)      (0.094)
                                                                    SD                    0.598***
                                                                                           (0.189)
                              Earnings (log)                     Mean        0.153***    0.156***
                                                                              (0.040)      (0.044)
                              Hours (log)                        Mean           0.605        0.639
                                                                              (0.516)      (0.544)
                              Hours (log)2                       Mean          -0.069       -0.067
                                                                              (0.078)      (0.081)
                              Tenure (log)                       Mean          -0.021       -0.037
                                                                              (0.047)      (0.053)
                              Apprentice                         Mean           0.251        0.228
                                                                              (0.342)      (0.424)
                              Household Assests (log)            Mean        0.069**       0.066*
                                                                              (0.033)      (0.038)
                              Male                               Mean          -0.074       -0.049
                                                                              (0.091)      (0.100)
                              Age                                Mean          -0.049       -0.047
                                                                              (0.031)      (0.030)
                              Age2 /100                          Mean           0.057        0.056
                                                                              (0.038)      (0.038)
                              Education                          Mean          -0.007        0.003
                                                                              (0.030)      (0.033)
                              Education2 /100                    Mean           0.005       -0.052
                                                                              (0.213)      (0.225)
                              Height (Z-score)                   Mean           0.002      0.0001
                                                                              (0.040)      (0.042)
                              Married                            Mean           0.121        0.140
                                                                              (0.091)      (0.101)
                              Household head                     Mean          -0.044       -0.068
                                                                              (0.086)      (0.096)
                              σ                                                           0.248***
                                                                                           (0.039)
                              Ethnicity Dummies                                Yes            Yes
                              City Dummies                                     Yes            Yes
                              Year Dummies                                     Yes            Yes
                              LL                                            -1034.690    -1029.193
                              Finite Sample AIC                               2.730          2.730
                              HQ IC                                           2.804          2.814
                              N                                                784            784
                              Individuals                                      784            249

Note: Estimation by Ordered Probit and Random Effects Random Parameter Ordered Probit. Following Dunne et al (1989),
the sample includes only long-survivors who are observed in every wave after �rst appearance. The dependent variable is Job
satisfaction which is an ordered variable with the following categories: 1. Very Dissatis�ed, 2. Dissatis�ed, 3. Neither Satis�ed
nor Dissatis�ed 4. Satis�ed 5. Very Satis�ed. The �ve labor market states are Self-no employees (those declaring themselves
self-employed, but with no employees); Self-with employees (those with employees); Informal salaried (employees in �rms of up
to 5 workers); Formal salaried (employees in �rms over 5 workers); and Civil (civil or public sector workers). Omitted category is
Formal salaried. Parameters for all job types- Self with employees Self-no employees, Informal Salaried and Civil are assumed to
follow a normal distribution. Estimates using 100 Halton draws. σ corresponds to random effect parameter which is estimated
assuming a random constant. Finite Sample AIC corresponds to AIC + 2M (M + 1)/(n − M − 1) and HQ IC corresponds to
(−2 log L + 2M log log n)/n where M is the number of parameters in the model. Thresholds nor reported. Standard errors in
parentheses. ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01.




                                                               44
                        Table 9: Sources of Premia and Heterogeneity: Autonomy, Job Characteristics
                        and Risk Aversion (RERPOP)

                                                         Autonomy          Occupation   and Sector       Risk Aversion
                                     Parameter        (1)          (2)        (3)           (4)          (5)        (6)
    Self-employed, employees           Mean        0.189*       0.198*     0.243***      0.335***       0.361      0.359
                                                   (0.110)      (0.110)     (0.085)       (0.098)     (0.231)    (0.232)
                                        SD        0.707***     0.704***    0.571***      0.578***      0.0002     0.004
                                                   (0.077)      (0.077)     (0.055)       (0.055)     (0.157)    (0.158)
    Self-employed, no employees        Mean         0.026        0.030       0.050         0.136       -0.155     -0.163
                                                   (0.093)      (0.093)     (0.075)       (0.091)     (0.168)    (0.169)
                                        SD        0.179***     0.185***    0.151***      0.154***     0.182**    0.178**
                                                   (0.040)      (0.040)     (0.029)       (0.029)     (0.074)    (0.074)
    Informal Salaried                  Mean         -0.052       -0.046      0.002         0.043       -0.195     -0.212
                                                   (0.099)      (0.099)     (0.085)       (0.086)     (0.192)    (0.195)
                                        SD           0.012        0.010      0.064         0.062        0.025      0.029
                                                   (0.063)      (0.064)     (0.056)       (0.056)     (0.134)    (0.133)
    Civil                              Mean          0.063        0.066      0.076         0.060        0.205      0.201
                                                   (0.132)      (0.132)     (0.110)       (0.113)     (0.227)    (0.227)
                                        SD        0.769***     0.768***    0.653***      0.647***     0.417**    0.415**
                                                   (0.107)      (0.107)     (0.091)       (0.091)     (0.204)    (0.204)
    Control (self-reported)            Mean                       0.046
                                                                (0.034)
    Wage & Manufacturing               Mean                                               0.060
                                                                                         (0.097)
    Self & Trade                       Mean                                               0.009
                                                                                         (0.056)
    Wage & not a manual laborer        Mean                                              0.176**
                                                                                         (0.085)
    Risk Aversion                      Mean                                                                        -0.004
                                                                                                                  (0.008)
    σ                                               0.006        0.006      0.092***     0.087***      0.102**    0.087*
                                                   (0.027)      (0.027)      (0.020)      (0.020)      (0.050)    (0.050)
    Ethnicity Dummies                                Yes          Yes          Yes          Yes          Yes        Yes
    City Dummies                                     Yes          Yes          Yes          Yes          Yes        Yes
    Year Dummies                                     Yes          Yes          Yes          Yes          Yes        Yes
    Other controls                                   Yes          Yes          Yes          Yes          Yes        Yes
    LL                                            -2177.824    -2176.940   -3597.253    -3595.069     -595.914   -595.719
    Finite Sample AIC                               2.566        2.566        2.578        2.579        2.612      2.616
    HQ IC                                           2.607        2.609        2.608        2.611        2.724      2.731
    N                                               1726         1726         2821         2821          487        487
    Individuals                                     1194         1194         1376         1376          197        197

Note: Estimation by Random Effects Random Parameter Ordered Probit (RERPOP). The dependent variable is Job satisfaction
which is an ordered variable with the following categories: 1. Very Dissatis�ed, 2. Dissatis�ed, 3. Neither Satis�ed nor
Dissatis�ed 4. Satis�ed, 5. Very Satis�ed. Sample common across all estimates and is limited by the fact that Life and
Financial satisfaction measures were collected only over 4 waves. The �ve labor market states are Self-no employees (those
declaring themselves self-employed, but with no employees); Self-with employees (those with employees); Informal salaried
(employees in �rms of up to 5 workers); Formal salaried (employees in �rms over 5 workers); and Civil (civil or public sector
workers). Omitted category is Formal salaried. Parameters for all job types- Self with employees Self-no employees, Informal
Salaried and Civil are assumed to follow a normal distribution. Omitted category is Formal employment. Estimates using
100 Halton draws. σ corresponds to random effect parameter which is estimated assuming a random constant. Controls in
both models correspond to Earning (log), Hours (log), Hours (log)2 , Tenure (log), Apprentice, Household Assets (log), Male,
Age, Age2 /100, Education, Education2 /100, Height (Z-score), Married, Household head, Ethnicity dummies and city dummies.
Finite Sample AIC corresponds to AIC + 2M (M + 1)/(n − M − 1) and HQ IC corresponds to (−2 log L + 2M log log n)/n where
M is the number of parameters in the model. Thresholds nor reported. Standard errors in parentheses. ∗ p < 0.1, ∗∗ p < 0.05,
∗∗∗ p < 0.01.




                                                              45
                    Figure 1: Distribution of Job Satisfaction by Sector (Model 3)




Note: The kernel densities correspond to the conditional mean of βik , that is, the �gure shows the kernel densities for the
distribution of estimates of E (βik |yi , xi ) across the sample. See Greene and Hensher (2010b) for more information on how to
obtain this estimation.




                                                             46
Figure 2: Distribution of job Satisfaction by Sector, Conditional on Remuneration (Model 6)




Note: The kernel densities correspond to the conditional mean of βik , that is, the �gure shows the kernel densities for the
distribution of estimates of E (βik |yi , xi ) across the sample. See Greene and Hensher (2010b) for more information on how to
obtain this estimation.




                                                             47
                                                Table 10: Combining Self-Employed

                                                               Parameter         (1)            (2)
                               Self-employed                     Mean           0.048          0.089
                                                                              (0.063)        (0.063)
                                                                  SD         0.187***       0.093***
                                                                              (0.024)        (0.025)
                               Informal Salaried                 Mean          -0.114         -0.013
                                                                              (0.073)        (0.073)
                                                                  SD         0.184***       0.165***
                                                                              (0.047)        (0.047)
                               Civil                             Mean           0.111          0.054
                                                                              (0.087)        (0.088)
                                                                  SD         0.359***       0.383***
                                                                              (0.069)        (0.070)
                               Earnings (log)                    Mean                       0.231***
                                                                                             (0.020)
                               Hours (log)                       Mean                        0.411*
                                                                                             (0.247)
                               Hours   (log)2                    Mean                         -0.058
                                                                                             (0.037)
                               Tenure (log)                      Mean                         -0.001
                                                                                             (0.023)
                               Apprentice                        Mean                          0.074
                                                                                             (0.132)
                               Household Assests (log)           Mean                       0.095***
                                                                                             (0.016)
                               Male                              Mean            0.035        -0.073
                                                                               (0.044)       (0.045)
                               Age                               Mean         -0.026**     -0.042***
                                                                               (0.013)       (0.013)
                               Age2 /100                         Mean           0.032*      0.050***
                                                                               (0.016)       (0.017)
                               Education                         Mean           -0.014        -0.002
                                                                               (0.014)       (0.015)
                               Education2 /100                   Mean         0.215**          0.025
                                                                               (0.097)       (0.099)
                               Height (Z-score)                  Mean         0.047**        0.036*
                                                                               (0.021)       (0.020)
                               Married                           Mean         0.104**          0.049
                                                                               (0.044)       (0.045)
                               Household head                    Mean          -0.076*        -0.042
                                                                               (0.042)       (0.043)
                               σ                                              0.245***      0.198***
                                                                               (0.019)       (0.019)
                               Ethnicity Dummies                                  Yes           Yes
                               City Dummies                                       Yes           Yes
                               Year Dummies                                       Yes           Yes
                               LL                                            -4223.207     -4138.240
                               Finite Sample AIC                                 2.625         2.576
                               HQ IC                                             2.645         2.601
                               N                                                 3242          3242
                               Individuals                                       1434          1434



                           Table 11: Share of Sector that Would not Prefer Formal Employment

                                                                 Model (1)     Model (2)
                                           Self-employed           60%           83%
                                           Informal Salaried       27%           47%
                                           Civil                   62%           56%

Note: Computed as the proportion of the population with a positive premium as 100 × Φ(βk /σk ), where Φ is the cumulative
standard normal distribution and βk and σk is the mean and the standard deviation of the coefficient as estimated in Table 10




                                                                 48