24
The Journal of Socio-Economics 37 (2008) 213–236 Happiness and health: Well-being among the self-employed Pernilla Andersson Swedish Institute for Social Research, Stockholm University, Sweden Received 13 April 2006; received in revised form 26 January 2007; accepted 15 March 2007 Abstract Is well-being greater among the self-employed than among wage-earners? In order to investigate this question, data from the Swedish Level-of-Living Survey for the 2 years 1991 and 2000 are used and six indicators of well-being are considered: (1) job satisfaction, (2) life satisfaction, (3) whether the job is stressful, (4) whether the job is mentally straining, (5) mental health problems, and (6) poor general health. Six logit models are estimated and to handle the possible selection of more satisfied individuals and individuals more able to handle stress into self-employment, conditional fixed-effects logit models are estimated for each of the outcomes. We find that self-employment leads to an increase in job satisfaction. We also find a positive correlation between self-employment and life satisfaction. There is some evidence that self- employment leads to more mental health problems, and that the self-employed are less likely to perceive their job as mentally straining. © 2007 Elsevier Inc. All rights reserved. JEL classification: J28; I31; I1 Keywords: Self-employment; Job satisfaction; Life satisfaction; Health 1. Introduction The interest among economists for job and life satisfaction in the population has increased over the last few years. Pioneering works on job satisfaction as an economic variable include Hamermesh (1977), Freeman (1978), and Borjas (1979). The classical economic view of happi- ness, or utility, has meant that economists have looked primarily on the relation between income Tel.: +46 8 163447. E-mail address: pernilla.andersson@sofi.su.se. 1053-5357/$ – see front matter © 2007 Elsevier Inc. All rights reserved. doi:10.1016/j.socec.2007.03.003

Happiness and health: Well-being among the self-employed

Embed Size (px)

Citation preview

Page 1: Happiness and health: Well-being among the self-employed

The Journal of Socio-Economics 37 (2008) 213–236

Happiness and health: Well-beingamong the self-employed

Pernilla Andersson ∗Swedish Institute for Social Research, Stockholm University, Sweden

Received 13 April 2006; received in revised form 26 January 2007; accepted 15 March 2007

Abstract

Is well-being greater among the self-employed than among wage-earners? In order to investigate thisquestion, data from the Swedish Level-of-Living Survey for the 2 years 1991 and 2000 are used and sixindicators of well-being are considered: (1) job satisfaction, (2) life satisfaction, (3) whether the job isstressful, (4) whether the job is mentally straining, (5) mental health problems, and (6) poor general health. Sixlogit models are estimated and to handle the possible selection of more satisfied individuals and individualsmore able to handle stress into self-employment, conditional fixed-effects logit models are estimated foreach of the outcomes. We find that self-employment leads to an increase in job satisfaction. We also finda positive correlation between self-employment and life satisfaction. There is some evidence that self-employment leads to more mental health problems, and that the self-employed are less likely to perceivetheir job as mentally straining.© 2007 Elsevier Inc. All rights reserved.

JEL classification: J28; I31; I1

Keywords: Self-employment; Job satisfaction; Life satisfaction; Health

1. Introduction

The interest among economists for job and life satisfaction in the population has increasedover the last few years. Pioneering works on job satisfaction as an economic variable includeHamermesh (1977), Freeman (1978), and Borjas (1979). The classical economic view of happi-ness, or utility, has meant that economists have looked primarily on the relation between income

∗ Tel.: +46 8 163447.E-mail address: [email protected].

1053-5357/$ – see front matter © 2007 Elsevier Inc. All rights reserved.doi:10.1016/j.socec.2007.03.003

Page 2: Happiness and health: Well-being among the self-employed

214 P. Andersson / The Journal of Socio-Economics 37 (2008) 213–236

and satisfaction (see, e.g., Clark and Oswald, 1996; Frey and Stutzer, 2002a,b; Blanchflower andOswald, 2003).

But issues related to job and life satisfaction and to the mental and physical health of theworkforce are important also for employers and policy makers. People who are more satisfiedwith their jobs might want to work harder and are less likely to quit, and people who are in goodhealth, both mentally and physically are more likely to be able to work. But a non-negligibleshare of the Swedish population has more or less severe mental health problems. It is estimatedthat between 20 and 40% of the Swedish population have mental health problems ranging fromsleeping problems and anxiety to psychoses (Folkhalsorapport, 2005).

Being self-employed and being a wage-earner is quite different as regards the working condi-tions, and as is argued below, we have reasons to believe that the happiness and health of thesetwo groups differ. Several earlier studies have for example shown that the self-employed are moresatisfied with their jobs than wage-earners are (Blanchflower and Oswald, 1998; Benz and Frey,2004; Blanchflower, 2004; Taylor, 2004). There is less evidence that life satisfaction is higheramong the self-employed. Benz and Frey (2004) identify certain factors that explain higher levelsof job satisfaction reported by the self-employed. These include being independent and one’sown boss, the absence of hierarchy, control over one’s own working hours, and over the effortexpended on the job.

On the contrary, it is assumed that the self-employed perceive their jobs as more stressful andmentally straining since they work longer hours, have less free time and more responsibility fortheir own jobs and incomes, as well in many cases for those of employees. For the same reasonsit is hypothesised that the self-employed have more mental health problems and a lower levelof general health than wage-earners. Several studies have shown that long working hours areassociated with health problems, for instance, physical health and tiredness (see, e.g., Van derHulst, 2003). Few studies have analysed the effect of self-employment on health. Blanchflower(2004) estimates the correlation between self-employment and different indicators of well-beingbut here one does not try to establish a causal link between them.

The purpose of the present paper is to analyse differences in well-being between wage-earnersand the self-employed in terms of six chosen indicators: (1) job satisfaction, (2) life satisfaction, (3)whether the job is stressful, (4) whether the job is mentally straining, (5) mental health problems,and (6) poor general health. Panel data from the Swedish Level-of-Living Survey (SLLS) for the2 years 1991 and 2000 (Erikson and Aberg, 1987; Fritzell and Lundberg, 1993) have been used.The focus is on estimating the causal effect of self-employment on these outcomes.

The main contribution of this paper is that the comparison between the self-employed andthe wage-earners regarding the non-pecuniary aspects of their occupation include not only jobsatisfaction and life satisfaction but also perceptions of the job as mentally straining and stressful,mental health, and general health.

A second contribution of the paper is that it tries to resolve the methodological concern aboutthe possible selection of happier and more satisfied individuals into self-employment in a slightlydifferent way than what has been done in earlier studies.1 The problem lies in how to estimate thecausal effect of self-employment on the outcomes. If there is a selection into self-employment wewould expect the self-employed to have been more satisfied with their jobs, even if they themselveshad not chosen to become self-employed. Also in the case of different health indicators and

1 See Clark and Oswald (2002) for a general discussion of problems associated with the analysis of people’s subjectivefeelings.

Page 3: Happiness and health: Well-being among the self-employed

P. Andersson / The Journal of Socio-Economics 37 (2008) 213–236 215

perceptions of the job as stressful or mentally straining, there is potential concern of the selectioninto self-employment. Those who choose to become self-employed may be those who like towork hard, who may see a demanding job as stimulating, or they may find it easier to handlestress and pressure at work than those who become wage-earners.

Since panel data are available, the traditional way in which we handle selection problems is toestimate a fixed-effects model in order to control for the unobserved heterogeneity. However, self-reported variables such as job and life satisfaction often appear as categorical variables in surveys.The number of response alternatives generally varies between 3 and 10. It is not straightforwardhow to handle this in a fixed-effects model. Various methods have been applied in earlier studies,namely random-effects ordered probit models (Taylor, 2004), fixed-effects models where thedependent variable is treated as a continuous variable (Benz and Frey, 2004; Blanchflower, 2004),and fixed-effects ordered logit models (D’Addio et al., 2004). Benz and Frey (2004) have alsoused a natural experiment approach.

I adopt a different estimation strategy compared with those previously used in this line ofresearch. The dependent variables are transformed into dichotomous variables and conditionalfixed-effects logit models are estimated to achieve a clearer interpretation of the differencesbetween wage-earners and the self-employed as regards well-being. The general idea of usingpanel data methods is that the effect of self-employment on job satisfaction and the other outcomes(life satisfaction, perceptions about the job, and health) is identified by individuals who, betweentwo time periods, have moved from non-self-employment to self-employment, or vice versa.

The main findings are that the self-employed are more satisfied with their jobs and this is nota result of selection. A strong positive correlation between life satisfaction and self-employmentis found but we cannot determine whether this is caused by selection or if it is a true causalrelationship. We also find that the self-employed are less likely to perceive their jobs as men-tally straining, a result contrary to our expectations. There is no evidence that the self-employedare more likely to perceive their jobs as stressful, nor is their general health poorer. However,we do find indications that self-employment increases mental health problems, in particular,tiredness.

The remainder of the paper is organised as follows. In Section 2 earlier research is summarised,in Section 3 the data set is described, in Section 4 the method and the empirical specification ofthe models are described, and in Section 5 sample means and changes in the dependent variablesare described and discussed. In Section 6 the results are presented and Section 7 summarises andconcludes the paper.

2. Earlier research

Blanchflower and Oswald (1998) used data from surveys performed in 1989 in the U.S., theU.K., and Germany to study self-employment and happiness. They found that many respondentsstate that they would like to be self-employed if they could. An implication of this is that those whoactually are self-employed should report a higher level of life and job satisfaction. As a measureof happiness, or overall utility, they use answers to a question on how satisfied or dissatisfied therespondents are with their jobs and a question on how satisfied they are with their lives as a whole.Self-employment turns out to have a positive significant effect on both these outcomes and theauthors draw the conclusion that the self-employed are more satisfied with their jobs and theirlives than wage-earners. Blanchflower (2000) draws the same conclusion based on the resultsof several job satisfaction equations estimated for 11 countries in Blanchflower and Freeman(1997). He concludes: “The self-employed are more satisfied with their jobs than are individuals

Page 4: Happiness and health: Well-being among the self-employed

216 P. Andersson / The Journal of Socio-Economics 37 (2008) 213–236

who work for somebody else” (Blanchflower, 2000, p. 21). Oswald (1997) too finds that reportedhappiness is high among the self-employed.

In a more recent paper, Blanchflower (2004), using data from the U.S. and Europe, finds thatthe self-employed report a higher level of job satisfaction than wage-earners.2 The same resultwas found by Taylor (2004) using the British Household Panel Survey (BHPS), and by Benz andFrey (2003, 2004) using the BHPS, the German Socioeconomic Panel Survey (GSOEP) and theSwiss Household Panel Survey. This result holds when fixed-effects models are estimated. Benzand Frey ask why the self-employed are happier with their jobs, and conclude that the higher levelof job satisfaction is due to independence at work and the absence of hierarchy. According to theseauthors, independence at work appears to have a value of its own. In the psychological literatureit is well known that independence is important to people’s well-being (Blanchflower, 2004).Benz and Frey show that when controls are included for evaluation of autonomy and evaluationof “work is interesting” in an ordered logit regression for job satisfaction, the coefficient for self-employment becomes insignificant. Using the Eurobarometer Survey (EBS), Blanchflower looksat the impact of self-employment on various indicators of mental health. The coefficient for theself-employed (with employees of their own) is positive and significantly different from zero forthe indicators “job is stressful”, “feels exhausted after work”, “being tired”, “being fed up withwork”, “loses sleep”, and “feels unhappy and depressed”.

So far, most results are related to the way the self-employed feel about their work. Blanchflower(2004) looks at eight different countries including Sweden and finds no significant effect onlife satisfaction from being self-employed. However, for 10 countries the coefficient for self-employment enters the regression with a positive sign and is significantly different from zero. Heconcludes that although the self-employed show signs of possessing a worse mental health thanwage-earners, they appear to be more satisfied with life.

3. Data

I use data from the Swedish Level-of-Living Survey, a survey that was conducted in 1968, 1974,1981, 1991, and 2000 (Erikson and Aberg, 1987; Fritzell and Lundberg, 1993). A representativesample of approximately 1/1000 of the adult Swedish population, in total about 5500 individuals,has been interviewed about their living and working conditions. Some of these individuals havebeen interviewed in several waves of the survey, so the interviews with them can be combinedto form a panel. The data used in the present study are restricted to the two latest waves, 1991and 2000, since the questions about job satisfaction, life satisfaction, and general health were notincluded until the wave of 1991. Only those who were working, either as wage-earners or as self-employed in both waves, are included. When individuals with missing information on some of thevariables are excluded, the sample consists of 1998 individuals. In 1991, 7.5% (149 individuals)were self-employed and in the year 2000, 10.4% (207 individuals) were self-employed. A majority(87.8%) were wage-earners in both years, 5.6% were self-employed in both waves, 4.7% switchedfrom being wage-earners in 1991 to being self-employed in 2000, and 1.8% switched from beingself-employed in 1991 to being wage-earners in 2000.

As noted in the Section 1, I study the relationship between self-employment and six differentindicators: (1) job satisfaction, (2) life satisfaction, (3) whether the job is stressful, (4) whether thejob is mentally straining, (5) mental health problems, and (6) poor general health. In the survey

2 The data used is the General Social Survey (GSS) for the U.S. and the Eurobarometer Survey for Europe.

Page 5: Happiness and health: Well-being among the self-employed

P. Andersson / The Journal of Socio-Economics 37 (2008) 213–236 217

Table 1Description of dependent variables

Dependent variables Description

Job satisfaction 1 if one is very satisfied with work, 0 otherwiseLife satisfaction 1 if one is satisfied with life most of the time, 0 otherwiseJob is stressful 1 if the job is perceived as stressful, 0 otherwiseJob is mentally straining 1 if the job is perceived as mentally straining, 0 otherwiseMental health problems 1 if one has had sleeping problems, been tired, been depressed, or anxious,

0 if no such problemsPoor general health 1 if general health is perceived to be poor or not so good, 0 if general

health is perceived to be good

questions are asked about these indicators and the response alternatives vary from two (job isstressful (yes/no) and job is mentally straining (yes/no)) to five (job satisfaction). The variableused to study mental health problems is an index based on information about whether an individualhas had problems with sleep, tiredness, anxiety, or depression, or if he or she has not had suchproblems. I have chosen to transform these indicators into dummy variables. One drawback ofdoing so is that it reduces the variation in the dependent variables. There are two reasons formaking the transformation. First, it means that the same method can be applied to the analysisof all outcomes, which simplifies the comparisons and interpretation of the results. Secondly,since we want to use the panel structure of the data, estimating logit models in the first stepallows us to estimate conditional fixed-effects models to control for unobserved heterogeneity.Ordered logit models have often been used in the base analysis in the earlier literature and all theinformation in the answers has then been used. However, there may be problems in estimatingfixed-effects models since fixed-effects ordered logit is not yet a common method. One exceptionin this context is D’Addio et al. (2004), where such an estimation method is suggested. In Taylor(2004), random-effects ordered probit models are estimated to deal with the selection problem,and in Benz and Frey (2004) ordinary least squares fixed-effects estimators are used for thesame purpose. Using these methods is not without problems, and there is a trade-off betweentransforming the variables in a suitable way, and using all variation in the dependent variables andusing a less suitable fixed-effects model. In Bjorklund (1985) and Winkelmann and Winkelmann(1998) the dependent variable, which is mental health problems, is transformed in the same wayas has been done here. Table 1 describes how the variables are defined and the wordings of thequestions are given in Appendix B.

In Section 1 we noted that the relation between happiness, or utility, and income has been studiedby economists, the hypothesis being that higher income leads to higher levels of happiness, orutility. In theory we might want to consider income as an important factor affecting both job andlife satisfaction for wage-earners as well as for the self-employed. If the self-employed receivehigher incomes than wage-earners and we also find them to be happier, this might just be a resultof receiving a higher income. However, it is not all that simple to put this into practice. The registerincomes reported by the self-employed and the wage-earners are not the same. Self-employed withprivate business report their “income of business activity” to the tax authorities and wage-earnersreceive a wage income. It is possible to add register information on incomes to the individuals inthe Level-of-Living Survey and this has been done. To have the same income measure for both theself-employed and wage-earners I have chosen total labour income, which is the sum of incomefrom business activity, wage income, and some social security payments (parents’ allowance andsickness benefits paid by the employer). Estimation of a simple income regression shows that

Page 6: Happiness and health: Well-being among the self-employed

218 P. Andersson / The Journal of Socio-Economics 37 (2008) 213–236

the self-employed have on average about 35% lower annual labour incomes (see Table A1 inAppendix A). A fixed-effects model also shows a significant decrease in the incomes for thosewho become self-employed. In Sweden, in general, we observe that the self-employed have lowerincomes than the wage-earners.3 So if the hypothesis is that the self-employed have higher joband life satisfaction but at the same time, they receive lower incomes, it is not certain that we wantto control for income. Including income as a control variable in the fixed-effects models is evenmore problematic. Since becoming self-employed for the average individual means a decreasein income, it can be hard to disentangle the effect of becoming self-employed from the effect ofreceiving a lower income on the outcomes. As a result, income is not included as a covariate inthe main analysis.4

4. Method and empirical specification

Since panel data are available, we can use a fixed-effects version of the logit model to consideran individual specific effect that is constant over time.5 This can be especially important whensurvey data are used since there is a possible anchoring problem, i.e. individuals differ in respectto the level at which they anchor their scales. When fixed-effects models are used, we calculatethe variation in the outcome within the group, i.e. the individual. We compare the outcome ofindividual i at time t with the outcome of the same individual at time t + 1. In this setting, theindividual-specific effect, ci, drops out of the equation. The estimated coefficients are then basedon intra-personal changes rather than on differences between individuals.

Chamberlain (1980, 1984) describes how fixed-effects models can be estimated when there is adichotomous dependent variable—the conditional fixed-effects logit model. Here we are interestedin estimating the effect of changes in the covariates on changes in the dependent variable. In atwo-period model we are interested only in cases where either yi,1 = 0 and yi,2 = 1 or yi,1 = 1 andyi,2 = 0, yi,1 is the outcome for individual i at time period 1 and yi,2 is the outcome for individual iat time period 2. This case can be described as yi,1 + yi,2 = 1. Let us define a new variable, wi, andlet wi = 1 if (yi,1,yi,2) = (0,1) and wi = 0 if (yi,1,yi,2) = (1,0). Following Maddala (1987) we candescribe the relation between the logit model and the fixed-effects logit model in a two-periodmodel as follows. Let the logit be

Pr(yit = 1) = exp(β′xit + ci)

1 + exp(β′xit + ci)(1)

Then

Pr(wi = 1) = 1

1 + exp(β′xi1 + ci)

exp(β′xi2 + ci)

1 + exp(β′xi2 + ci)(2)

and

Pr(wi = 0) = exp(β′xi1 + ci)

1 + exp(β′xi1 + ci)

1

1 + exp(β′xi2 + ci)(3)

3 Andersson (2006).4 Estimating the logit models with a control for income does not change the results (see Tables A2.1 and A2.2 in

Appendix A).5 See Long (1997) for a description of the logit model and Chamberlain (1980, 1984), Maddala (1987) for descriptions

of the fixed-effects conditional logit model.

Page 7: Happiness and health: Well-being among the self-employed

P. Andersson / The Journal of Socio-Economics 37 (2008) 213–236 219

Since these two cases are mutually exclusive

Pr(wi = 1|yi,1 + yi,2 = 1) = Pr(wi = 1)

(Pr(wi = 0) + Pr(wi = 1))= exp(β′(xi1 − xi2))

1 + exp(β′(xi1 − xi2))(4)

and

Pr(wi = 0|yi,1 + yi,2 = 1) = Pr(wi = 0)

(Pr(wi = 0) + Pr(wi = 1))= 1

1 + exp(β′(xi1 − xi2))(5)

the individual specific effect has now disappeared from the equation, and we have dealt withthe problem of time-invariant omitted variables or unobserved heterogeneity. In this model wecondition on yi,1 + yi,2 = 1, i.e. that there has been a change in the dependent variable between thetwo time periods. Individuals who do not change state are not included in the regressions, sincethey contribute zero to the likelihood function (Maddala, 1987).

The logit and the fixed-effects logit are not the easiest models to interpret quantitatively, andthe characteristics of the fixed-effects logit leads to a drastic drop in the number of observationsused to estimate the model. As a sensitivity check, I, therefore, estimate all models using a linearprobability model both on the cross-sectional data and on the panel data.

To describe the empirical specification of the model, let SE = 1 if an individual is self-employedand SE = 0 if not, T = 1 if year = 2000 and T = 0 if year = 1991. Further, X is a vector of additionalcontrols such as age, gender, education, and marital status.6 The unobserved heterogeneity, or theindividual-specific effect, which is assumed to be correlated with the outcome, and the decisionto become self-employed is denoted with ci. In the logit model this term appears in the errorterm and so, if ci is correlated with self-employment, as we expect it to be, this leads to biasedestimates.

The time dummy, T, included in the logit regressions captures the effect of changes in societyand on the labour market between 1991 and 2000. We estimate two different specifications

Model(i) Pr(yi,t = 1|SE, T ) = exp(β1SEi,t + β2Tt + ci)

1 + exp(β1SEi,t + β2Tt + ci)(6)

Model (ii) Pr(yi,t = 1|SE, T, X) = exp(β1SEi,t + β2Tt + δXi,t + ci)

1 + exp(β1SEi,t + β2Tt + δXi,t + ci)(7)

This is all fairly straightforward: β1 measures the association between being self-employedand the outcome and β2 is the difference in the outcome between the 2 years.

In the fixed-effects version of the model we estimate the same specifications as in the logitmodel with one difference: in the fixed-effects model the time dummy is excluded, since thechange in this variable is the same for all individuals. In the conditional fixed-effects logit modelthe probability of having reported a change from 0 to 1 (wi = 1) in the dependent variable iscalculated as:

Pr(wi = 1|yi,1 + yi,2 = 1) = exp(β(xi,2 − xi,1))

1 + exp(β(xi,2 − xi,1))(8)

To be more specific, Model (i) can be written as:

Pr(wi = 1|yi,1 + yi,2 = 1) = exp(β1(SEi,2000 − SEi,1991))

1 + exp(β1(SEi,2000 − SEi,1991))(9)

6 For a full description of covariates included in X, see text below Tables 5.1–5.6.

Page 8: Happiness and health: Well-being among the self-employed

220 P. Andersson / The Journal of Socio-Economics 37 (2008) 213–236

where SE = 1 means that the individual is self-employed and SE = 0 that the individual is a wage-earner. β1 gives an estimate of the causal effect of self-employment on the outcomes. Model (ii)is also estimated using the conditional fixed-effects estimation procedure but with time-varyingcovariates included in the X vector only.

In identifying the effect of self-employment on the different outcomes information is used onlyon individuals whose answer in the dependent variable changed between the years concerned. InSection 1 we discussed the type of individual characteristics that are assumed to be included in theunobserved heterogeneity. We assume that the unobserved heterogeneity captures such personaltraits as are correlated with the way an individual perceives his or her life and environment ingeneral. Unobserved heterogeneity can also include the ability to cope with stress and pressure,whether an individual has a positive or negative view of work and life, or has a positive andoptimistic attitude in everyday life. If there is a correlation between this unobserved factor, thedependent variable, and the decision to become self-employed, the estimates in the logit modelare biased, and it cannot be said that we have estimated the “true” effect of self-employment onthe outcome.

5. Descriptive statistics

5.1. Sample means

Table 2 provides descriptive statistics for the sample. For each year the population is dividedinto wage-earners and self-employed. The self-employed are older, a smaller share of them arewomen, a larger share are married, a larger share live in rural areas, and on average they havehad fewer years of education. An examination of health variables, perceptions regarding work,happiness, and job characteristics reveals some interesting differences. In the 1991 sample thereappear to be no differences between the self-employed and wage-earners. The share of the self-employed who said that they were very satisfied with their work is not significantly differentfrom the wage-earners’ share. Nor is there a difference in the share of those who find their jobsmentally straining. When these questions were asked again in 2000 the answers were different. Asignificantly lower share of the self-employed compared to the wage-earners say that their jobs arementally straining. This is particularly interesting in light of the fact that the share who said so roseamong the wage-earners while it fell among the self-employed. The same pattern appears regardingjob satisfaction and life satisfaction: in 1991 there were no significant differences between wage-earners and the self-employed, but in 2000 a larger share of the self-employed said that they werevery satisfied with their jobs and for most of the time felt satisfied with their lives (Table2).

One difference that does not change between the years is that on average the self-employedwork significantly more hours per week than wage-earners do. However, the question put to thetwo groups is not the same.7

The annual incomes presented in Table 2 are collected from register data. This variable is usedsince there is no income measure gathered from the self-employed in the survey. The incomesin 1991 are expressed in the prices of 2000. The self-employed have on average lower annual

7 To wage-earners the question is “How many hours per week is your ordinary working time?” To the self-employedthe question is “How many hours do you usually work in the firm per week on average over a year?” Since “ordinaryworking time” referrers to the working time stated in the contract and not to the actual hours worked, the comparison canbe misleading. However, even if overtime is added to the working time for wage-earners, it is not likely that they work onaverage as many hours as the self-employed do.

Page 9: Happiness and health: Well-being among the self-employed

P. Andersson / The Journal of Socio-Economics 37 (2008) 213–236 221

Table 2Descriptive statistics (%)

Wage-earnersin 1991

Self-employedin 1991

Wage-earnersin 2000

Self-employedin 2000

Socio-economic characteristicsAge (years) 37.1 41.2*** 46.2 48.0**

Women 49.1 26.2*** 49.7 27.5***

Unmarried 21.1 13.4*** 12.8 6.3***

Divorced 4.4 3.4 6.6 6.3Widow 0.5 0.0*** 1.9 0.5**

Married 74.0 83.2*** 78.6 86.9***

Children living at home 1.1 1.2 0.86 0.82

Place of residenceStockholm 15.5 18.1 16.1 17.4Gothenburg 7.7 6.7 7.4 6.8Malmoe 4.5 1.3*** 4.1 2.4City > 30 000 inhabitants 22.1 16.1* 19.2 15.0City < 30 000 inhabitants 19.9 17.4 18.4 15.9Rural area 30.3 40.3** 34.8 42.5**

Native 93.1 87.2** 92.7 92.3Education (years) 11.7 10.9*** 12.2 11.7**

Measures of well-beingLife satisfaction 61.4 64.4 59.1 70.5***

Job satisfaction 42.8 48.3 31.4 49.8***

Job is stressful 68.0 72.5 72.5 75.4Job is mentally straining 48.6 52.3 53.7 46.4**

Mental health problems 22.2 20.1 33.0 29.9Poor general health 9.7 11.4 19.6 18.4Feeling overstrained 6.0 9.4 12.4 13.5

Job characteristicsWeekly working time (h) 37.3 49.2*** 37.6 47.6***

Annual labour income(SEK)a (S.D.)

1 98 546 (97 296) 1 78 129** (118 246) 2 53 535 (163 610) 2 01 151*** (143 275)

Satisfied with the wage 53.0 69.8*** 45.1 63.3***

Feeling control over life 75.1 79.9 75.4 86.5***

Number of individuals 1 849 149 1791 207

a The average annual exchange rate in 2000: 1D = 8.5 SEK.* Differences between the mean values for wage-earners and the self-employed in 1991 and 2000 are significantly

different from each other at the 10% level of significance.** Differences between the mean values for wage-earners and the self-employed in 1991 and 2000 are significantly

different from each other at the 5% level of significance.*** Differences between the mean values for wage-earners and the self-employed in 1991 and 2000 are significantlydifferent from each other at the 1% level of significance.

income but the spread is larger than among wage-earners. This is the same pattern we find inregister data for the whole Swedish population (Andersson, 2006). It can, however, be a bitmisleading to compare register incomes for the two groups since it is often assumed that theincomes reported by the self-employed to the tax authorities are understated.8 In spite of a lower

8 See Engstrom and Holmlund (2006) for an analysis of the underreporting of incomes among the self-employed inSweden.

Page 10: Happiness and health: Well-being among the self-employed

222 P. Andersson / The Journal of Socio-Economics 37 (2008) 213–236

registered income, a larger share of the self-employed stated that they are satisfied with theirincome.

The survey includes a question about whether the respondents feel that they are in control oftheir everyday lives. A slightly larger share of the self-employed says that they feel they are incontrol of their lives.

There are very interesting differences, primarily for wage-earners, between the years. Theshare of the population of wage-earners who gave a “negative” answer to the outcome variablesis much higher in 2000 than in 1991; a higher share stated that their general health was poor, ahigher share had mental health problems, and a higher share stated that their job was stressful.The share of those who were very satisfied with the job also declined and we can observe a slightdecrease in the share who said they were satisfied with their lives for most of the time.

To some extent the same pattern appears for the self-employed, but with some importantexceptions. The share of the self-employed who stated that their jobs were mentally straining issmaller in 2000 than in 1991 while a slightly higher percentage said in 2000 that they were verysatisfied with life in general and with their work compared to in 1990.

An explanation for the differences between 1991 and 2000 could be that time pressure hasincreased in the society.9 In general, the greater time pressure is assumed to have a negative effecton people’s health, both mental and physical. Since the changes over time for the self-employedand the wage-earners are different, it is possible that there are differences in their ability to handlethe increased pace of life, and it is possible that those who are able to handle it are overrepresentedamong the self-employed.

Even though the purpose of this paper is not to make a thorough investigation of the relationshipbetween the different outcomes and whether high life satisfaction causes a better health or if thecausality goes in the opposite direction, it can be of interest to describe the correlations. This isdone separately for the self-employed and wage-earners. All correlations have the expected sings:a positive correlation between the “positive” outcomes (job satisfaction and life satisfaction), andbetween the four “negative” outcomes (job is stressful, job is mentally straining, mental healthproblems, and poor general health) and a negative correlation between the positive and the negativeoutcomes. We see, for example, that having mental health problems and a poor general health isnegatively correlated with life satisfaction as is also noted in Gerdtham and Johannesson (2001).An exception to this pattern is that there is a very small positive correlation for wage-earnersbetween life satisfaction and feeling that the job is stressful mentally straining. It is, however,only the correlation between feeling that the job is mentally straining and life satisfaction that issignificantly different from zero.

Almost all correlations are stronger among the self-employed. For example, the correlationbetween job satisfaction and life satisfaction is 0.326 for the self-employed and 0.234 for wage-earners and the correlation between feeling that the job is mentally straining and life satisfactionis −0.136 for the self-employed and 0.033 among wage-earners. Here, there is even a differencein the sign of the correlation. Having mental health problems appears to be correlated with theother outcomes pretty much in the same way for wage-earners and the self-employed. For bothgroups we see, for example, a high positive correlation between having mental health problemsand perceiving one’s general health to be poor.

Labour income, which is defined as total income from work (self-employment, paid employ-ment, parents allowance, and sickness benefits paid by the employer), is positively correlated

9 See, for example, Garhammer (2003).

Page 11: Happiness and health: Well-being among the self-employed

P. Andersson / The Journal of Socio-Economics 37 (2008) 213–236 223

with life and job satisfaction but also with feeling that the job is stressful and mentally straining.People with jobs that are stressful and mentally straining do tend to receive a higher income. Itis noteworthy that this correlation is much stronger for the self-employed. There is, as can beexpected, a negative correlation between income and the two health variables, but we cannot sayanything about the direction of causality (Tables 3a and 3b).10

5.2. Changes in the dependent variables

Table 4 shows the changes observed in the data between 1991 and 2000 for four groups: thosewho are self-employed in both periods, those who are wage-earners in 1991 and self-employed in2000, those who are self-employed in 1991 and wage-earners in 2000, and those who are wage-earners in both 1991 and 2000. Table 4 shows the shares for which the dependent variable is onein each year and for each group. The asterisks indicate whether the shares in 1991 and 2000 aresignificantly different from each other.

Regardless of the year, the self-employed work on average 10 more hours per week then wage-earners. Those who became self-employed increased their average working time from 37.8 to48.1 h per week, while those who left self-employment reduced their working time by approxi-mately the same number of hours. In the group who remained wage-earners, a significantly largershare in 2000 than in 1991 stated that they felt that the job was stressful and mentally straining.None of the other groups experienced such a development. Another difference is that those whobecame self-employed and those who continued to be wage-earners experienced a significantincrease in the share with mental health problems which, for wage-earners, can be correlated withthe fact that they felt that pressure on the job increased.

In earlier studies it has been argued that job satisfaction increases for those who become self-employed. Our data show an interesting pattern. Among individuals who were wage-earners in1991 and self-employed in 2000, the share of those who were very satisfied with their jobs in 2000is not significantly larger than the share of those who reported the same in 1991, i.e. when theywere wage-earners. During the same period, job satisfaction decreased for those who continuedto be wage-earners. The share of those who were very satisfied with their jobs fell from 42.5%in 1991 to 31.2% in 2000. Among those who left self-employment, the share of those who werevery satisfied fell from 55.6% in 1991 to 38.6% in 2000. In sum, Table 4 shows that the group thatbecame self-employed has experienced different changes compared with those who continuedto be wage-earners. One reason could be that those who became self-employed were differentin some way, but a comparison with the answers in 1991 indicates that those who later becameself-employed are fairly similar to those who continued to be wage-earners.

6. Results

The main results are presented in Tables 5a–5f. In all tables the coefficients are presented.11

Also note that the number of individuals in the logit and the fixed-effects models differs since inthe fixed-effects models we only use information on individuals who have reported a change inthe dependent variable between 1991 and 2000.

10 See, for example, Smith (1999) for an extensive discussion about the relationship between income and health.11 The odds ratios can be calculated by taking exp(β) for the group that has the value 1 of the covariate in question and

exp(−β) for the group that has the value 0 of the covariate.

Page 12: Happiness and health: Well-being among the self-employed

224P.A

ndersson/T

heJournalofSocio-E

conomics

37(2008)

213–236

Table 3aCorrelations between the dependent variables, time, and labour income for the self-employed

Self-employed (N = 356) Job satisfaction Life satisfaction Job is stressful Job is mentallystraining

Mental healthproblems

Poor generalhealth

Year 2000

Life satisfaction 0.326*

Job is stressful −0.126* −0.103*

Job is mentally straining −0.163* −0.136* 0.282*

Mental health problems −0.183* −0.117* 0.099* 0.193*

Poor general health −0.187* −0.206* 0.164* 0.142* 0.316*

Year 2000 0.014 0.065 0.032 −0.059 0.111* 0.095*

Labour income 0.096* 0.037 0.209* 0.181* −0.088* −0.055 0.085

* Correlation is significantly different from 0 at least at the 10% level of significance.

Page 13: Happiness and health: Well-being among the self-employed

P.Andersson

/The

JournalofSocio-Econom

ics37

(2008)213–236

225

Table 3bCorrelations between the dependent variables, time, and labour income for wage-earners

Wage-earners (N = 3 640) Job satisfaction Life satisfaction Job is stressful Job is mentallystraining

Mental healthproblems

Poor generalhealth

Year 2000

Life satisfaction 0.234*

Job is stressful −0.071* 0.020Job is mentally straining −0.034* 0.033* 0.247*

Mental health problems −0.111* −0.105* 0.083* 0.106*

Poor general health −0.106* −0.147* 0.034* 0.016 0.290*

Year 2000 −0.118* −0.024 0.049* 0.051* 0.121* 0.141*

Labour income 0.035* 0.076* 0.054* 0.103* −0.052* −0.064* 0.201*

* Correlation is significantly different from 0 at least at the 10% level of significance.

Page 14: Happiness and health: Well-being among the self-employed

226 P. Andersson / The Journal of Socio-Economics 37 (2008) 213–236

Table 4Changes in hours worked and the dependent variables between 1991 and 2000 for different groups

Hours worked and dependentvariables

Self-employedboth in 1991and 2000

Wage-earner in 1991and self-employed in2000

Self-employed in 1991and wage-earner in2000

Wage-earnerboth in 1991and 2000

Hours worked in 1991 (per week) 49.1 37.8 49.1 37.3Hours worked in 2000 47.2 48.1*** 38.8*** 37.6Very satisfied with the job in 1991 46.0 47.9 55.6 42.5Very satisfied with the job in 2000 42.5 58.5 38.9 31.2***

Satisfied with life in general in 1991 63.7 67.0 66.7 61.1Satisfied with life in general in 2000 69.0 72.3 55.6 59.2Job is stressful in 1991 75.2 74.5 63.9 67.6Job is stressful in 2000 76.1 74.5 69.4 72.5***

Job is mentally straining in 1991 50.4 58.5 58.3 48.1Job is mentally straining in 2000 43.4 50.0 61.1 53.6***

Mental health problems in 1991 17.7 20.2 27.8 22.3Mental health problems in 2000 25.7 35.1** 36.1 32.9***

Poor general health in 1991 11.5 11.7 11.1 9.6Poor general health in 2000 22.1** 13.8 22.2 19.5***

Number of observations 113 94 36 1755

** Differences between the mean values in 1991 are significantly different from the mean values in 2000 for each groupat the 5% level of significance.*** Differences between the mean values in 1991 are significantly different from the mean values in 2000 for each groupat the 1% level of significance.

In the logit models for job satisfaction, we find the coefficient for self-employed to be positiveand significantly different from 0. One matter of concern is that the individuals who were dissatis-fied with their jobs in 1991 might be more likely to switch to self-employment. It has been shownin other studies that individuals who are less satisfied with their jobs are more likely to quit (Clark,2001; Kristensen and Westergard-Nielsen, 2004). If this is so, then we would expect individualswho were wage-earners in 1991 and self-employed in 2000 to be less satisfied with their jobs in1991 than the group of individuals who were wage-earners in both years. From Table 4 we seethat this is not the case.

In the fixed-effects model where we control for a number of individual characteristics (edu-cation, marital status, place of residence, having children at home) the difference between

Table 5aJob satisfaction

Logit (i) Logit (ii) FE (i) FE (ii)

Self-employed 0.537 (0.112)*** 0.621 (0.115)*** 0.528 (0.263)** 0.582 (0.268)**

Year 2000 −0.440 (0.066)*** −0.494 (0.074)*** – –Other controls No Yes No YesObservations (individuals) 3 996 (1998) 3 996 (1998) 1 456 (728) 1 456 (728)Number of parameters 2 15 1 11LR Chi2 64.24 100.63 4.18 35.91

Notes: Coefficients, standard errors given in parentheses. Model (ii) also includes controls for age, gender, place of birth(not FE), education, marital status, place of residence, and having children at home.** Significant at 5%.

*** Significant at 1%.

Page 15: Happiness and health: Well-being among the self-employed

P. Andersson / The Journal of Socio-Economics 37 (2008) 213–236 227

Table 5bLife satisfaction

Logit (i) Logit (ii) FE (i) FE (ii)

Self-employed 0.341 (0.119)*** 0.384 (0.122)*** 0.343 (0.279) 0.346 (0.282)Year 2000 −0.066 (0.065) −0.169 (0.074)** − –Other controls No Yes No YesObservations (individuals) 3 996 (1998) 3 996 (1998) 1 450 (725) 1 450 (725)Number of parameters 2 15 1 11LR Chi2 9.26 115.97 1.54 7.50

Notes: Coefficients, standard errors given in parentheses. Model (ii) also includes controls for age, gender, place of birth(not FE), education, marital status, place of residence, and having children at home.** Significant at 5%.

*** Significant at 1%.

Table 5cJob is stressful

Logit (i) Logit (ii) FE (i) FE (ii)

Self-employed 0.179 (0.127) 0.232 (0.130)* −0.105 (0.325) −0.188 (0.334)Year 2000 0.210 (0.070)*** 0.352 (0.079)*** – –Other controls No Yes No YesObservations (individuals) 3 996 (1998) 3 996 (1998) 1 262 (631) 1 262 (631)Number of parameters 2 15 1 11LR chi2 11.60 66.44 0.11 17.63

Notes: Coefficients, standard errors given in parentheses. Model (ii) also includes controls for age, gender, place of birth(not FE), education, marital status, place of residence, and having children at home.

* Significant at 10%.*** Significant at 1%.

self-employed and wage-earners is still significantly different from 0. The conclusion is thenthat those who become self-employed are more likely to report an increase in job satisfaction.On a basis of this result we can say that self-employment appears to have a causal effect on jobsatisfaction.

In the logit models we get the same result for life satisfaction as for job satisfaction: the self-employed are more likely to state that they are satisfied with their lives for most of the time than

Table 5dJob is mentally straining

Logit (i) Logit (ii) FE (i) FE (ii)

Self-employed −0.105 (0.111) 0.051 (0.117) −0.642 (0.391) −0.682 (0.395)*

Year 2000 0.165 (0.063)*** 0.160 (0.073)** − –Other controls No Yes No YesObservations (individuals) 3 996 (1998) 3 996 (1998) 1 182 (591) 1 182 (591)Number of parameters 2 15 1 11LR Chi2 7.46 248.46 2.84 16.72

Notes: Coefficients, standard errors given in parentheses. Model (ii) also includes controls for age, gender, place of birth(not FE), education, marital status, place of residence, and having children at home.

* Significant at 10%.** Significant at 5%.

*** Significant at 1%.

Page 16: Happiness and health: Well-being among the self-employed

228 P. Andersson / The Journal of Socio-Economics 37 (2008) 213–236

Table 5eMental health problems

Logit (i) Logit (iii) FE (i) FE (iii)

Self-employed −0.136 (0.128) 0.028 (0.132) 0.580 (0.334)* 0.493 (0.340)Year 2000 0.543 (0.072)*** 0.731 (0.082)*** – –Other controls No Yes No YesObservations (individuals) 3 996 (1998) 3 996 (1998) 1 220 (610) 1 220 (610)Number of parameters 2 15 1 11LR chi2 57.99 187.59 3.15 21.69

Notes: Coefficients, standard errors given in parentheses. Model (ii) also includes controls for age, gender, place of birth(not FE), education, marital status, place of residence, and having children at home.

* Significant at 10%.*** Significant at 1%.

are wage-earners. In the specification where we include the additional covariates (age, gender,education, marital status, place of residence, having children at home, and place of birth) thecoefficient for the time dummy is negative and significant which means that the individuals in theworking population were less likely in 2000 to state that they were satisfied with their lives formost of the time than they were in 1991.

The coefficient in the simple fixed-effects model (no additional covariates) is approximately thesame as the estimate in the corresponding logit model, but the standard errors are larger. Thus whilethe estimate in the logit is significantly different from 0, the estimate in the fixed-effects model isnot. Although we do not get an estimate in the fixed-effects model which is significantly differentfrom 0, we can nonetheless say that the result about the self-employed being more satisfied withtheir lives in general is not a result of a selection of happier people into self-employment. Thereason for getting an insignificant estimate in the fixed-effects model is rather that there is notenough variation in the outcome, that is to say, individuals do not tend to change their perceptionsabout their lives from one year to another.

The third outcome is perceptions of the job as stressful. In the logit estimates of the full modelwe find that there is a significant difference between wage-earners and the self-employed: self-employed individuals are more likely to state that their job is stressful. Note that we have notcontrolled for working time in this model. If this is done (result omitted), then self-employmentdoes not have a significant impact on the outcome. The estimate for the time dummy is positiveand significantly different from 0 in all models, implying that workers in general are more likely

Table 5fPoor general health

Logit (i) Logit (iii) FE (i) FE (iii)

Self-employed 0.002 (0.156) −0.118 (0.161) −0.167 (0.410) −0.253 (0.446)Year 2000 0.799 (0.094)*** 0.599 (0.107)*** – –Other controls No Yes No YesObservations (individuals) 3 996 (1998) 3 996 (1998) 706 (353) 706 (353)Number of parameters 2 15 1 11LR chi2 75.81 195.06 0.17 54.97

Notes: Coefficients, standard errors given in parentheses. Model (ii) also includes controls for age, gender, place of birth(not FE), education, marital status, place of residence, and having children at home.*** Significant at 1%.

Page 17: Happiness and health: Well-being among the self-employed

P. Andersson / The Journal of Socio-Economics 37 (2008) 213–236 229

to state that the job is stressful in 2000 than they were in 1991. In the fixed-effects models we findno significant effect from self-employment on the likelihood of perceiving the job as stressful.

The result regarding perceptions of the job as mentally straining is similar to that found regard-ing perception of the job as stressful: there is no strong evidence that the self-employed are morelikely to find their job mentally straining. In the full fixed-effects model we find the coefficientfor self-employed to be negative and significantly different from 0. This result is contrary towhat we expected. We argued in Section 1 that being self-employed is more likely to be a men-tally challenging occupation. A possible explanation for this somewhat surprising finding is thatindividuals who left their employment with a firm to become self-employed may have done sobecause they perceived their job at the firm as mentally straining. In comparison, self-employmentis experienced as less mentally straining. One way of checking the plausibility of this explanationis to look at Table 4 again. We see that a larger share of those who were wage-earners in 1991 andself-employed in 2000 said in 1991 that they felt their job was mentally straining. In the year 2000,this share had dropped while the corresponding change for individuals who were wage-earners atboth times was increased. Thus, this does seem to be a reasonable explanation for the result.

The fifth outcome is if one has mental health problems or not. In the logit models we find nodifference between wage-earners and the self-employed and we see the same pattern over time asbefore: workers are more likely to state that they have mental health problems in 2000 than in 1991.

In the simple fixed-effects model (no covariates added) we find a positive and significant effectfrom becoming self-employed on the probability of changing from not having mental healthproblems to having such problems. Hence, there is some evidence that self-employment appearsto increase mental health problems, as expected.12 In fact, it can be argued that the pattern of anon-significant effect in the logit model and a positive and significant effect in the fixed-effectsmodel is consistent with the underlying selection mechanism proposed earlier. Assume that self-employed individuals are better equipped than wage-earners to handle stress, pressure, and longdays of hard work. Add to this the fact that the self-employed work more and probably havemore responsibility. The non-effect in the logit model can then be seen as a result of the negativeeffect of more work and responsibility on mental health on the one hand, and the greater abilityto cope with this on the other hand cancel each other out. When instead in the fixed-effects modelwe compare the mental health of the same person at two different points in time, we find thatthose who become self-employed are more likely to say that they perceive their mental health ashaving deteriorated. However, the inclusion of covariates (children living at home, marital status,place of residence, and education) in the fixed-effects model reduces the estimate and increasesthe standard errors so that the significant effect of self-employment on mental health problemsdisappears.13

The analysis of general health does not reveal any significant differences between wage-earnersand the self-employed, either in the logit models or in the fixed-effects models.

The result that self-employment increases mental health problems and that it decreases theprobability of finding the job mentally straining might not be so intuitively appealing. It is hardto say exactly why we get this result since we have seen that these two outcomes are positivelycorrelated with each other. One explanation for why self-employment decreases the risk of finding

12 Mental health problems are a combination of “being tired”, “having sleeping-problems”, “feeling anxious”, and “beingdepressed” (see Appendix B). A separate analysis of each of these four indicators shows that the effect arise primarilyfrom the self-employed being more likely to feel tired.13 Inclusion of the covariates – children living at home, marital status, place of residence, and education – one at a time

reveals that place of residence contributes to the insignificant coefficient for self-employed.

Page 18: Happiness and health: Well-being among the self-employed

230 P. Andersson / The Journal of Socio-Economics 37 (2008) 213–236

the job mentally straining might be that the self-employed enjoy their work more and people whoare more satisfied with their work are less likely to fell that the job is mentally straining. We findthat when controlling for job satisfaction the estimated effect decreases and becomes insignificant,hence the effect of self-employment can partly be attributed to an increase in job satisfaction whenbecoming self-employed.

When controlling for job satisfaction in the regression for mental health problems we find thatthe positive effect of self-employment actually increases and once again, this result mainly arisefrom that self-employment increases tiredness.

It seems as higher job satisfaction can have a protective effect on how you perceive yourown work but it cannot protect you from mental health problems. Even if the mental health isself-reported it is a more objective measure than if one feels that the job is mentally straining.

One drawback of presenting the estimated coefficients in the logit and the fixed-effects logitmodels is that the coefficients cannot be given a quantitative interpretation, and as was mentionedabove, the two types of models are not based on exactly the same sample since we in the fixed-effects model only use individuals for which we observe a change in the outcomes betweenthe years. As a comparison, model (ii) is re-estimated using a linear probability model (seeTables A3.1–A3.3 in Appendix A). The results from the linear probability models are in all cases,except one, qualitatively the same as in the logit and fixed-effects models. We find a differenceregarding the effect of becoming self-employed on an increase in the risk of perceiving the job asmentally straining. In the fixed-effects logit we got a negative and significant estimate while weget an insignificant estimate in the linear probability model.

Regarding the size of the effect, we can see from the linear probability models that the proba-bility for the self-employed stating that they are very satisfied with their job is about 15 percentagepoints higher compared to wage-earners. In the fixed-effects version of the linear probability modelthe estimated effect is around 13 percentage points. The correlation between self-employment andlife satisfaction is slightly lower—8.6 percentage points.

7. Summary and conclusion

In this paper well-being among the self-employed is analysed using data from the SwedishLevel-of-Living Survey for the years 1991 and 2000. Six indicators of well-being have beenconsidered: (1) job satisfaction, (2) life satisfaction, (3) whether the job is stressful, (4) whetherthe job is mentally straining, (5) mental health problems, and (6) poor general health.

The expectation is that there are differences between wage-earners and the self-employedregarding these outcomes. For the two measures of satisfaction it is believed that the self-employedare more satisfied. Being independent and acting as one’s own boss, the absence of hierarchy,control over working hours, and the effort put into the job are all factors that are assumed to explainhigher levels of job satisfaction. If someone is satisfied with their job, then they are also assumedto be more satisfied with life in general. It is assumed that the self-employed perceive their jobsas more stressful and mentally straining since they work more and have more responsibility.For the same reasons it is assumed that the self-employed have more mental health problems andpoorer general health than wage-earners do. Earlier research has shown that there is an associationbetween long working hours and health problems.

The empirical strategy is to estimate both logit models and conditional fixed-effects models.Fixed-effects versions of the models are estimated, since we believe that there is a selection of acertain type of individuals into self-employment. Regarding the two “positive” outcomes we areconcerned that in the logit models we may overestimate the positive effect of being self-employed

Page 19: Happiness and health: Well-being among the self-employed

P. Andersson / The Journal of Socio-Economics 37 (2008) 213–236 231

if a selection of happier and more satisfied individuals into self-employment does occur. Whencontrolling for the individual-specific effect in the fixed-effects models we expect the estimate todecrease and perhaps even become insignificant. For the “negative outcomes”, that is perceptionsof the job as stressful and mentally straining and health problems, the selection mechanism isassumed to be that individuals who become self-employed are more able to handle stress andpressure, and hence their health will not be so negatively affected by hard work and stress.

The main findings of the paper are that those who become self-employed are more likely toreport an increase in job satisfaction. This is a result in line with earlier research. Regardinglife satisfaction we find a strong positive correlation, i.e. the self-employed appear to be moresatisfied with their lives than wage-earners. In the fixed-effects models the estimates are no longersignificantly different from 0, so we cannot interpret this relation in causal terms.

The self-employed are neither more nor less likely than wage-earners to perceive their jobsas stressful. Also, individuals who become self-employed are not more likely to experience adeterioration in their general health than those who are wage-earners in both years.

Self-employment does appear to have an effect on mental health problems: those who becomeself-employed are more likely to experience a deterioration in mental health compared to indi-viduals who do not change their occupational status. This result is, however, not robust to theinclusion of all additional covariates.

Regarding perceptions of the job as mentally straining the results are contrary to those weexpected: those who switch to self-employment appear to be less likely to perceive that their jobhas become mentally straining. This finding can perhaps be seen as a result of the self-employedenjoying more freedom and self-determination at work. Under these circumstances, long workinghours and responsibility are perceived as a burden but as a challenge and as something thatenhances motivation.

A conclusion to be drawn from our results is that if wage-earners enjoy more of the indepen-dence, freedom, and lack of hierarchy that the self-employed do, they might feel more satisfiedwith the job. It is also important to remember that the self-employed are more positively disposedtowards work, even though they work on average 10 h more per week and lower incomes.

So is well-being higher among the self-employed? The only really robust result is that self-employment increases job satisfaction. For the other outcomes we get mixed results dependingon the specification of the model and method used. But our results point towards the fact thatwell-being is at least not lower among the self-employed, this in spite of that they work more andreceive lower incomes.

Acknowledgements

I would like to thank Anders Bjorklund, Rickard Eriksson, Mats Hammarstedt, MagnusJohannesson, Mikael Lindahl, Magnus Lofstrom, Eskil Wadensjo, seminar participants at theSwedish Institute for Social Research, the Institute for Labour Market Policy Evaluation, and twoanonymous referees for valuable comments on earlier versions of this paper.

Appendix A. Appendix A

Table A1.Tables A2.1 and A2.2.Tables A3.1–A3.3.

Page 20: Happiness and health: Well-being among the self-employed

232 P. Andersson / The Journal of Socio-Economics 37 (2008) 213–236

Table A1Income regressions for the self-employed and wage-earners

Annual labour income Annual labour incomeOLS FE

Self-employed −0.359 (0.044)*** −0.299 (0.073)***

Constant 10.470 (0.121)*** 12.602 (1.861)***

Individual characteristics* Yes YesObservations 3 967 3 967Number of individuals 1.996 1.996R-squared 0.29 0.20

Notes: Individuals characteristics include: age, gender, education, marital status, children, place or residence, place ofbirth. Robust standard errors given in parentheses.

* Significant at 10%.*** Significant at 1%.

Table A2.1Logit model with control for annual labour income

Job satisfaction[Logit (ii)]

Life satisfaction[Logit (ii)]

Job is stressful[Logit (ii)]

Job is mentallystraining [Logit (ii)]

Self-employed 0.707 (0.117)*** 0.486 (0.125)*** 0.351 (0.133)*** 0.152 (0.119)Annual labour

income/10.0000.015 (0.003)*** 0.018 (0.004)*** 0.020 (0.004)*** 0.019 (0.004)***

Year 2000 −0.533 (0.075)*** −0.215 (0.075)*** 0.306 (0.080)*** 0.114 (0.074)Other controls Yes Yes Yes YesObservations 3 995 3 995 3 995 3 995Number of individuals 1 997 1 997 1 997 1 997LR chi2 121.77 140.56 94.02 275.30

Notes: Coefficients, standard errors given in parentheses. Individuals characteristics include: age, gender, place of birth(not in the fixed-effects model), education, marital status, children, and place or residence.*** Significant at 1%.

Table A2.2Logit model with control for annual labour income

Mental health problems [Logit (ii)] Poor general health [Logit (ii)]

Self-employed −0.029 (0.134) −0.304 (0.167)*

Annual labour income/10.000 −0.010 (0.004)** −0.030 (0.006)***

Year 2000 0.757 (0.083)*** 0.670 (0.108)***

Other controls Yes YesObservations 3 995 3 995Number of individuals 1 997 1 997LR chi2 194.81 225.60

* Significant at 10%.** Significant at 5%.

*** Significant at 1%.

Page 21: Happiness and health: Well-being among the self-employed

P. Andersson / The Journal of Socio-Economics 37 (2008) 213–236 233

Table A3.1Linear probaility models

Job satisfaction Life satisfaction

OLS (ii) Fixed-effects (ii) OLS (ii) Fixed-effects (ii)

Self-employed 0.147 (0.028)*** 0.131 (0.053)** 0.086 (0.026)*** 0.070 (0.056)Year 2000 −0.114 (0.017)*** – −0.039 (0.017)** –Other controls Yes Yes Yes YesObservations 3 996 3 996 3 996 3 996Number of individuals 1 998 1 998 1 998 1 998R-squared 0.03 0.02 0.03 0.00

Notes: Individuals characteristics include: age, gender, place of birth (not in the fixed-effects model), education, maritalstatus, children, place or residence. Robust standard errors given in parentheses.** Significant at 5%.

*** Significant at 1%.

Table A3.2Linear probability models

Job is stressful Job is mentally straining

OLS (ii) Fixed-effects (ii) OLS (ii) Fixed-effects (ii)

Self-employed 0.046 (0.025)* −0.027 (0.049) 0.012 (0.028) −0.075 (0.048)Year 2000 0.071 (0.016)*** – 0.038 (0.017)** –Other controls Yes Yes Yes YesObservations 3 996 3 996 3 996 3 996Number of individuals 1 998 1 998 1 998 1 998R-squared 0.02 0.01 0.06 0.01

* Significant at 10%.** Significant at 5%.

*** Significant at 1%.

Table A3.3Linear probability models

Mental health problems Poor general health

OLS (ii) Fixed-effects (ii) OLS (ii) Fixed-effects (ii)

Self-employed 0.005 (0.024) 0.073 (0.048) −0.015 (0.020) −0.025 (0.037)Year 2000 0.140 (0.016)*** – 0.069 (0.012)*** –Other controls Yes Yes Yes YesObservations 3 996 3 996 3 996 3 996Number of individuals 1 998 1 998 1 998 1 998R-squared 0.05 0.01 0.05 0.02

*** Significant at 1%.

Page 22: Happiness and health: Well-being among the self-employed

234 P. Andersson / The Journal of Socio-Economics 37 (2008) 213–236

Appendix B. Dependent variables—the questions

B.1. Job satisfaction

We have now asked you a number of questions about your work. In general, how satisfied areyou with your present job?

1. Very satisfied2. Fairly satisfied3. Neither satisfied nor dissatisfied4. Fairly dissatisfied5. Very dissatisfied

B.2. Life satisfaction

Do you generally feel that your life is a source of personal satisfaction?

1. Yes, most of the time2. Yes, sometimes3. No

B.3. Job is stressful

Is your job stressful?

1. Yes2. No

B.4. Job is mentally straining

Is your job mentally straining?

1. Yes2. No

B.5. Mental health

The variable used in the analysis is constructed from the following four questions:During the last 12 months, have you suffered from tiredness?

1. No2. Yes, slightly3. Yes, seriously

Page 23: Happiness and health: Well-being among the self-employed

P. Andersson / The Journal of Socio-Economics 37 (2008) 213–236 235

During the last 12 months, have you had difficulties in sleeping?

1. No2. Yes, slight3. Yes, serious

During the last 12 months, have you had nervous problems (anxiety worries)?

1. No2. Yes, slight3. Yes, serious

During the last 12 months, have you suffered depression?

1. No2. Yes, slight3. Yes, serious

Mental health is considered to be “normal” if the respondent has answered “No” to all theabove questions or if the respondent has had slight problems with tiredness or slight sleepingproblems. In all other cases the individual is considered to have had mental health problems.

B.6. General health

How do you judge the state of your general health? Is it . . .

1. Good2. Poor3. Something in between

References

Andersson, P., 2006. Self-employed immigrants in Sweden: are they as successful as natives? In: Four Essays in Self-Employment. Swedish Institute for Social Research, Stockholm University (Dissertation Series No. 69).

Benz, M., Frey, B.S., 2003. Being Independent is a Great Thing: Subjective Evaluations of Self-employment and Hierarchy(Working Paper No. 135). Institute for Empirical research in Economics, University of Zurich.

Benz, M., Frey, B.S., 2004. Being independent raises happiness at work. Swedish Economic Policy Review 12, 97–138.Bjorklund, A, 1985. Unemployment and mental health: some evidence from panel data. Journal of Human Resources 20,

469–483.Blanchflower, D.G., 2000. Self-employment in OECD countries. Labour Economics 7, 471–505.Blanchflower, D.G., 2004. Self-employment: more may not be better. Swedish Economic Policy Review 11 (2), 15–73.Blanchflower, D.G., Freeman, R., 1997. The attitudinal legacy of communist labor relations. Industrial and Labor Relations

Review 50, 438–459.Blanchflower, D.G., Oswald, A., 1998. What makes an entrepreneur? Journal of Labor Economics 16, 26–60.Blanchflower, D.G., Oswald, A., 2003. Well-being over time in Britain and the USA. Journal of Public Economics 88,

1359–1386.Borjas, G.J., 1979. Job satisfaction, wages, and unions. Journal of Human Resources 14, 21–40.Chamberlain, G., 1980. Analysis of covariance with qualitative data. Review of Economic Studies 47, 225–238.Chamberlain, G., 1984. Panel data. In: Griliches, Z., Intriligator, M. (Eds.), Handbook of Econometrics, vol. 2. North-

Holland, Amsterdam.

Page 24: Happiness and health: Well-being among the self-employed

236 P. Andersson / The Journal of Socio-Economics 37 (2008) 213–236

Clark, A, 2001. What really matters in a job? Hedonic measurement using quit data. Labour Economics 8, 223–242.Clark, A., Oswald, A., 1996. Satisfaction and comparison income. Journal of Public Economics 61, 359–381.Clark, A., Oswald, A., 2002. Well-Being in Panels. Working Paper at the Department of Economics, University of Warwick,

UK.D’Addio, A.C., Eriksson, T.T., Fritjers, P., 2004. An analysis of the determinants of job satisfaction when individuals’

baseline satisfaction levels may differ. In: Paper presented at the EALE conference, Lisbon.Engstrom, P., Holmlund, B., 2006. Tax evasion and self-employment in a high-tax country: evidence from Sweden

(Working Paper 2006:12). Uppsala University, Department of Economics.Erikson, R., Aberg, R., 1987. Welfare in transition—living conditions in Sweden 1968–1981. Clarendon Press, Oxford.Freeman, R., 1978. Job satisfaction as an economic variable. American Economic Review 68, 135–141.Frey, B., Stutzer, A., 2002a. Happiness and Economics: How the Economy and Institutions Affect Well-Being. Princeton

University Press, Princeton.Frey, B., Stutzer, A., 2002b. What can economists learn from happiness research? Journal of Economic Literature 40,

402–435.Fritzell, J., Lundberg, O., 1993. Ett forlorat eller forlovat artionde? Valfardsutvecklingen mellan 1981 och 1991. Institutet

for Social Forskning, Stockholm.Garhammer, M., 2003. Pace of Life and Enjoyment of Life. Paper Presented at the Conference The Paradoxes of Happiness

in Economics. University of Milano-Bicocca.Gerdtham, U., Johannesson, M., 2001. The relationship between happiness, health and socio-economic factors; results

based on Swedish micro data. Journal of Socioeconomics 30, 553–557.Hamermesh, D., 1977. Economic aspects of job satisfaction. In: Orely, C., Oates, W.E. (Eds.), Ashenfelter, Essays in

Labor Market Analysis. Wiley, New York.Van der Hulst, M., 2003. Long work hours and health. Scandinavian Journal of Work Environment and Health 29, 171–188.Kristensen, N., Westergard-Nielsen, N., 2004. Does Low Job Satisfaction Lead to Job Mobility? (IZA Discussion Paper

No. 1026).Long, S.J., 1997. Regression Models for Categorical and Limited Dependent Variables. Sage Publications, Thousand

Lakes, California.Maddala, G.S., 1987. Limited dependent variable models using panel data. Journal of Human Resources 22, 307–338.Oswald, A., 1997. Happiness and economic performance. Economic Journal 107, 1815–1831.Smith, J.P., 1999. Healthy bodies and thick wallets: The dual relations between health and economic status. Journal of

Economic Perspectives 3 (2), 145–166.Socialstyrelsen, 2005. Folkhalsorapport 2005.Taylor, M., 2004. Self-employment in Britain: when, who and why? Swedish Economic Policy Review 12 (2), 139–173.Winkelmann, L., Winkelmann, R., 1998. Why are the unemployed so unhappy? Evidence from panel data. Economica

65, 1–15.