Do individuals try to maximize general satisfaction?
Paul Frijters*
Faculty of Economics and Econometrics, Tinbergen Institute Amsterdam, University of Amsterdam, Roetersstraat 11, 1018 Amsterdam, Netherlands
Received 4 December 1998; received in revised form 10 September 1999; accepted 1 December 1999
Abstract
The hypothesis that individuals try to maximize their life-satisfaction is analyzed. The approach was to derive empirically testable predictions as to the relationships between in-tentions, actions, importance weights, and satisfaction levels that would be consistent with the hypothesis, and to test these predictions on a Russian and a German panel data set. The respondents investigated were more likely to intend to change those areas they are unsatis®ed with in this period, were more likely to actually have changed those areas they were unsatis®ed with in the last period, and tended to ®nd the areas of their lives they were dissatis®ed with less important. The relationships were not very strong though and were more reliable for the German data set than for the Russian data set. The ®ndings therefore give only limited support to the hypothesis examined. Ó 2000 Elsevier Science B.V. All rights reserved.
PsycINFO classi®cation:2340; 3120
JEL classi®cation:D00; D63; I30
Keywords:Satisfaction; Memory and anticipation; Choice behaviour; Panel data analysis www.elsevier.com/locate/joep
*Present address: Faculty of Economics, Free University Amsterdam, De Boelelaan 1105, 1081 HV
Amsterdam, The Netherlands. Tel.: +31-20-4446155; fax: +31-20-4446005; web.: http://www.econ.vu.nl/ medewerkers/pfrijters.
E-mail address:[email protected] (P. Frijters).
1. Introduction
In this paper, we investigate whether the choice behavior of individuals is consistent with the idea that they try to maximize self-reported levels of general satisfaction. Our purpose is to assess the possible usefulness of sat-isfaction questions for economics in general and policy questions in partic-ular. If individuals try to maximize self-reported satisfaction levels, then a satisfaction level corresponds to the economic notion of an ordinal utility level. Moreover, as satisfaction levels also have normative connotations in the classical utilitarian sense, they may be used to make normative statements about policies.
In this section, we use the psychological literature on satisfaction levels to build a descriptive model which explains how individuals answer satisfaction questions. We then derive two predictions as to the verbal and choice be-havior that would be consistent with the idea that individualstrytomaximize their general satisfaction level.
In Section 2, the predictions are tested on large Russian and German panel data sets with parametric and semi-parametric techniques. The main theo-retical prediction is that individuals will want to make changes in those areas of their lives that they are unsatis®ed with. This will be investigated with the help of partial satisfaction levels, which are the levels of satisfaction that individuals enjoy in speci®c areas of their lives. We investigate whether a low level of satisfaction in a particular area of life is correlated with the plan to change current conditions in that area, for several partial satisfactions of individuals, namely satisfaction with marriage, satisfaction with the current job, and satisfaction with current housing conditions.
We then study panel data sets to see whether low satisfaction in a speci®c area of life in a previous period has led to changes in those areas of life in this period. Finally, we investigate whether individuals tend to ®nd those areas of life important which gave them a lot of satisfaction in the past, since one of the easiest ways for an individual to be satis®ed in general is to concentrate on those areas with which he is satis®ed.
1.1. Current ®ndings
In the psychological literature there are several theories as to the meaning of self-reported satisfaction levels.1 Rather than discussing them all one by one, the interpretation used by Kahneman (1998) is followed because it is intuitively attractive and lends itself well for hypothesis building.
Like other vertebrates, humans have speci®c brain centers which contin-uously evaluate current experiences so as to allow them to make decisions about the continuation of current activities and to decide on their response to experiences (see Shizgal, 1998). These ``instant evaluations'', which in the human case are self-reported on a good/bad scale 2are termed ``momentary satisfaction levels'' (MSLs). In the conception used by Kahneman, which goes back to at least Bentham (1798) and Edgeworth (1881), self-reported levels of satisfaction are an aggregate of these MSLs. In its simplest form, satisfaction with life, also termed general satisfaction, is a weighted average of the MSLs over the periodT1 toT2:
GSi
Z T2
T1
mit MSLi;tMSLit
f gdt
and
Z T2
T1
mit MSLi;tdt1;
where GSi denotes the satisfaction with life as a whole as reported by indi-viduali, andm denotes a memory function which weights each momentary satisfaction level of the individual at timet, denoted by MSLit. The memory function depends ontand on the whole stream of MSLs, which is denoted by MSLi:The idea that general satisfaction is a weighted average of momentary satisfactions is re¯ected by the condition that the memory weightsmit add up to one. This way of arriving at a self-reported general satisfaction level is illustrated in Fig. 1, where we have taken m to be a constant and hence show the case where all MSLs have equal weights.
1See e.g., Veenhoven (1996), Diener, Suh and Oichi (1997), Van Praag and Frijters (1999) and the
chapters in Kahneman, Diener and Schwarz (1998) for overviews.
2Dierent studies use dierent wordings (good/bad, negative/positive, favorable/unfavorable, etc.)
General satisfaction equals the average level of momentary satisfactions fromT1toT2and equals the shaded area as time weights are equal to 1 in this example. In principle, the method allows for the idea that individuals include all the MSLs they ever hadandanticipations of MSLs in the future into their general satisfaction level. This way of conceptualizing satisfaction levels also allows for an interpretation of partial satisfaction levels. A partial satisfac-tion level is also a weighted average of MSLs but each partial satisfacsatisfac-tion has its own weight function: the moments an individual counts as relevant for a particular domain have a higher memory weight than the moments an in-dividual does not deem relevant for that domain. Some moments may be relevant for many domains and individuals may have other ideas about the boundaries of domains.
Three issues are elaborated, keeping in mind that we want to use the framework to assess whether individuals use satisfaction levels to make choices: how do individual circumstances aect MSLs, how do individuals construct the memory weightsmit, and how do individuals evaluate possible future events? Below, some empirical results on each of these questions will be presented and discussed. In each case, the literature will be used to for-mulate falsi®able hypotheses as to the behavior which would be consistent with the idea that individuals try to maximize their general satisfaction levels. It has been found that the level of momentary satisfaction individuals derive from an experience depends on a reference position, which in turn depends on previous and anticipated experiences. Put simply, the ``higher'' the reference position, i.e., the higher the level of anticipations, of an indi-vidual, the higher the outcome must be in order to achieve a certain mo-mentary satisfaction level. Because the reference position adapts with new
experiences, this ®nding is known best under the name adaptation theory. 3 The expectations an individual has of future experiences will depend on the ability of an individual to obtain favorable experiences. This in turn is de-termined by external constraints and individual abilities. In the terminology of the goal persuit literature (e.g. Gollwitzer, 1993; Heckhausen & Schultz, 1995), an individual will be dissatis®ed when outcomes are beneath what he believes to be within his ``primary control'': for example, an individual's expectation of his future income depends on the availability of work and on the ability of the individual to ®nd work and perform complicated tasks. If the realized experiences are better than expected, the individual will not ex-pect to be able to improve on these circumstances. If the realized experiences are worse than expected, an individual will hope to improve them by changing his circumstances. This leads to the hypothesis thatindividuals will intend to change the areas of their lives with which they are unsatis®ed in this period and will have changed those areas of their lives they were unsatis®ed with in the last period. We split this hypothesis in two parts in the results section. First, we consider whether theintentionsof the individual are related to low satisfaction levels with respect to the area involved. Secondly, we look at whether we can explain some actualchoicesmade by individuals by looking at the satisfaction values in the previous period.
The memory weightmitof an MSL has been found to depend on its relative position in a stream of MSLs (Kahneman, Fredrickson, Schreiber & Redelmeier, 1993). If the MSL of a speci®c moment is particularly pleasant or unpleasant relative to what has preceded or followed in a certain time period, the MSL of that moment will have a disproportionately large in¯u-ence on self-reported satisfaction levels. Extreme experiin¯u-ences are remembered better and, because remembering is a form of consumption, have a large in¯uence on subsequent MSLs. It has also been found to be the case that the last MSLs in a period are of particular importance for self-reported levels of satisfaction. This ®nding has been termed thepeak±end ruleby Kahneman et al. (1993). To clarify, suppose that in a medical operation without anaes-thetic, the pain (negative momentary satisfaction) experienced ®rst increases, reaches a peak, and then decreases. The evaluation that an individual gives of
3There is a whole host of theories which consider the idea that the reference position can be formed by
this operation can then be well-predicted by the level of pain experienced at the peak and at the end: the pain at other moments has hardly any percep-tible in¯uence on the evaluation.
An important point to watch out for in empirical analyses is that these memory weights are very easily manipulated by the researcher (see e.g. Schwarz, 1995, for an extensive discussion of this issue): the MSLs asso-ciated with the experiences an individual is reminded of just before an-swering satisfaction questions, are more important for that reported satisfaction level. As a demonstration of this bias, we report the results of a simple OLS-analysis. The data come from a sample of Russians in 1995 and will be described later in detail. The endogenous variable is the answer to the question ``how satis®ed are you with the ®nancial situation of your household on a scale from 1 to 10, where 1 stands for very unsatis®ed and 10 for very satis®ed''. The ®rst exogenous variable is the household income from all activities before tax. The second is the log-mean of the Income Evaluation Question and roughly equals the ®nancial needs of the family (see Appendix A). Now, the ®nancial satisfaction question happens to have been posed twice in the same interview. The ®rst time, it was posed before the exogenous variable-questions and the second time it was posed after. Regressing both endogenous variables on the same exogenous variables yields:
Looking only at the dierences in the coecients, we see from these OLS regressions that the eects of the exogenous variables were some 20% greater when the question on ®nancial satisfaction was asked immediately after these questions than when the question was asked before. Also, the empirical ®t has increased substantially, witness the large increase in R2. As these par-ticular exogenous variables, income and the ®nancial needs of a family (l), are major elements in the life of the respondent so that he will often be re-minded of them anyway, the eect of posing these questions just before the
The importance of the order of questions for ®nancial satisfaction* Financial
satisfaction 1
t-value Financial
satisfaction 2
t-value
ln (Income) 1.60 8.2 1.90 22.3
l )1.39 6.4 )1.62 17.1
N 1442 1442
R2 0.04 0.26
satisfaction question will probably be a lot less than for questions on oth-erwise less important and less remembered subjects. The 20% bias found here may be seen as a lower bound.
Not only a questionnaire can trigger memory: individuals have limited control over what they remember. Selective memory eects are known from the theory of cognitive dissonance (Festinger, 1957) which shows that indi-viduals are more likely to remember things if they ®t in with their pre-con-ceived ideas. It has also for long been known that individuals use their memory to explain events and occurrences in a way bene®cial to their own circumstances. For example, Pettigrew (1979) showed that successful indi-viduals attributed success in general topersonalfactors or factors under their control, whereas unsuccessful individuals attributed their lack of success to bad luck or other factors outside their control. Both groups interpreted re-ality in such a way that it promoted their own satisfaction and remembered those facts bene®cial to their point of view.
The possibility of a selective memory leads us to expect that individuals increase the memory weight of more pleasant events in their lives and reduce the memory weight of unpleasant events (repress them). This leads to the hypothesis that the importance of an area of one's life will increase if the
individual was more satis®ed with this area in previous periods. In other
Predictions of individuals of the satisfaction value of future events are prone to human de®ciencies. Firstly, individuals seem unable to anticipate changes in their reference position. As a result, they fail to anticipate that they will value something higher when they possess it. In Festinger (1957) experiment for instance, the subjective desirability of a consumer good as reported by test persons increased markedly after an individual had chosen that consumer object: an object that was purchased appeared more beautiful and more useful than it had before the purchase. The ®nding that individuals fail to anticipate before a purchase their change in evaluation after a pur-chase is known in economics as theendowment effect(Thaler, 1980), and has been replicated in many psychological choice-experiments (e.g. Kahneman, Knetsch & Thaler, 1991; Loewenstein & Adler, 1995; Simonson, 1990; Kahneman & Schnell, 1992).4
Another problem individuals face when they anticipate future outcomes is that they overestimate small probabilities and underestimate large proba-bilities, as demonstrated by Kahneman and Tversky (1979), Van de Stadt, Antonides and Van Praag (1984) and Tversky and Fox (1995). This helps to explain gambling, insurance and tournaments (see Frijters, 1998). The point of these two ®ndings for this paper is that it is hard to infer from the satis-faction effect of choices whether individuals were trying to maximize their satisfaction levels or not. Individuals may be unsuccessful in maximizing their general satisfaction because they fail to take a change of reference position into account and they cannot deal with small probabilities. We can therefore only check whether individuals' intentions and choices are consis-tent with the idea that theytry to maximize general satisfaction.
2. Data, methods and results
The ®rst panel data set used is the Russian National Panel data set, from which we used the ®rst two waves, collected in 1993 and 1994. This panel data set was designed to track intentions and actions of individuals over time and hence contains information for several areas of individuals' lives about the intentions that individuals have to change aspects of their lives in the
4It has been argued that this eect is evolutionary driven because individuals who have learned to value
future, the level of satisfaction that individuals currently attain within an area of their lives and whether any actions have been taken since the last in-terview in that area. With this panel data set we can hence test for several areas of individuals' lives whether they are more likely to intend to change aspects of their lives if they are more unsatis®ed with that area currently, and whether they have actually changed an aspect of their life they were unsatis®ed with last year. The correspondence between the intentions in 1993 and the actions in 1994 is not complete however. It is for instance not asked in 1994 whether an individual followed up on his intentions to change his marital status in 1993. All that is known is whether the marital status actually changed, but this can be the result of exogenous in¯uences such as the partner's actions. Hence the areas for which intentions are available do not perfectly overlap with the areas for which follow-up ac-tions are known. For some areas (such as satisfaction with social contacts), neither is known.
The second panel data set used is the German Socio-Economic Panel (GSOEP), described in Wagner, Burkhause and Behringer (1993)5 of which we used the 1992 and 1993 West-German waves. This panel data set was not designed to track intentions and actions over time. Therefore we have to use `loose' questions which were posed here and there in the questionnaires as indicators of whether individuals have an intention to change an area of their life or have changed an area of their life. This means there will be virtually no overlap between intentions and actual choices. The GSOEP panel does however have extensive information on socio-economic variables of indi-viduals, such as income, demographic composition, and education. Like the Russian panel data set, the GSOEP also has extensive information on the general satisfaction levels of the respondents and on satisfaction levels with particular areas of their lives.
What is missing for both data sets are questions that probe the personality of the individual involved. As a result, we will have to make do with de-mographic variables, such as age and gender, as additional explanatory variables for intentions and changes, apart from satisfaction variables.
5The GSOEP is a longitudinal household survey sponsored by the Deutsche Forschungsgemeinschaft
2.1. Descriptive statistics
For both data sets used, the sample averages and variances of some im-portant variables are given in Table 1.
The precise de®nitions of these and other variables are given in the ap-pendix. The most glaring dierence between the two data sets is that the respondents in the GSOEP are generally more satis®ed with all aspects of their lives (especially their ®nances), are younger, slightly more often male and better educated than their Russian counterparts.
As to the analyses, reweighting measures were used initially for both data sets to see whether sample-selection biased the results, but as they did not in¯uence the results signi®cantly, the results shown are on unweighted samples. A general discussion of selection issues and characteristics of the Russian data set can be found in (Frijters, 1999, Ch. 2).
2.2. The intention to change in Russia and Germany
In the Russian panel data set we have information available on the in-tentions of individuals to change their circumstances in three spheres of their lives, namely the intention of individuals to change their family conditions,
Table 1
Descriptive statistics of the Russian data set 1992 and the GSOEP 1993
Descriptive statistics of: The Russian data set 1992 The GSOEP 1993
Average S.D. Average S.D.
ln (monthly income) 10.9 0.8 7.9 1.7
Education 12.8 4.4 12.1 2.4
Male 0.37 0.4 0.43 0.5
Age 44 12 39 12
l 11.4 0.8
Life satisfaction 5.1 2.5 6.9 1.9
Health satisfaction 6.7 2.3
Marital satisfaction 7.7 2.7
Housing satisfaction 5.4 3.1 7.2 2.2
Job satisfaction 6.2 2.7 7.1 2.0
Social satisfaction 6.9 2.5
Leisure satisfaction 6.6 2.4
Financial satisfaction 3.0 2.2 6.2 2.3
Work condition satisfaction 5.4 2.7
Environmental satisfaction 5.5 2.2
their housing conditions and their job conditions. In Table 2, we report the eect of various factors on the probability that an individual intends to make speci®c changes to an aspect of his life. The dependent variable equals one if an individual indicated a speci®c intention to change an area of his life, and equals zero otherwise. We include variables denoting the ability of the in-dividual to change an area of his life, such as income, age, education, the ®nancial needs of the household (denoted by l) and gender as explanatory variables. The ®nancial need of the household here refers to the income the household claims to need to label that income level as ``not good, not bad''. We also include variables denoting the current satisfaction levels of the in-dividual and the relative place of the partial satisfaction level involved. We include two dummies, one that denotes whether the partial satisfaction in-volved is the maximum satisfaction level of all the partial satisfaction levels reported, and a second which denotes whether the partial satisfaction level equals the minimum partial satisfaction level as an indication of the relative place of a partial satisfaction level in the set of partial satisfactions.
Table 2 shows the results of a Probit analysis. A maximum score estimator was also used, which is a semi-parametric method which makes fewer as-sumptions on the error-distribution (see Horowitz, 1992, 1993). Because the results of the maximum score estimation are not substantially dierent from
Table 2
Concrete intentions to change an aspect of one's life by personal action in Russiaa
Family situation Housing situation Job situation
Probit t-value Probit t-value Probit t-value
Circumstances
ln (income) 0.06 0.9 0.22 4.4 )0.07 1.0
Education 0.02 1.8 0.01 2.0 0.01 0.9
Male )0.12 1.7 0.03 0.5 0.26 3.3 Satisfaction with this sphere )0.30 5.1 )0.26 5.1 )0.44 7.2 Maximum satisfaction? )0.003 0.0 )0.41 4.8 )0.11 1.0 Minimum satisfaction? )0.21 1.4 )0.23 2.7 0.01 0.0
ÿ2 Log L 1685.2 2845.3 1365.4
Number ofyiequal to 1 446 994 311
N 1646 2522 1693
aA polynomial age-pro®le was tried for all intentions. It was retained only if signi®cant. Constants are not
those of the Probit analyses however, the maximum score results are not shown but do suggest some robustness of the results.
The explanatory variables have been grouped into individual circum-stances and individual satisfaction variables. Looking at the individual cir-cumstances, we ®nd that age is by far the most important. 6 A higher age decreases the intention to change, in the case of family life even quadratically so (from 20 years of age onwards, the older the respondent gets, the in-creasingly unlikely becomes the intention of making changes in family life). As change brings ®xed costs, it may be expected that younger individuals have longer to retrieve these costs. Another explanation for the strong neg-ative in¯uence of age on the intention to change is that older respondents have alreadymade all the changes they believe could increase their general satisfaction.
Another individual circumstance, income, plays a less signi®cant role in the intention to change. Only the intention to change one's housing condi-tions is signi®cantly increased by a higher income. As changes in housing are likely to be costly, this was to be expected. The fact that higher income is negatively related to the intention to change one's job conditions is also understandable, as those with a high income will not want to change jobs. On the other hand, a high level of ®nancial need aects the intention to change. Individuals with relatively high ®nancial needs may look for a higher paying job. Gender is relevant for increasing the intention to change job-condi-tions.7 A possible explanation is that men are more ambitious in the workplace than women. In short, the eects of individual circumstances on the intention to change, seems quite reasonable.
As to the in¯uence of satisfaction variables, the most important factor in all three areas of satisfaction is the absolute satisfaction level of the particular sphere involved. The relative place of the partial satisfaction amongst all reported partial satisfaction levels is less important. It therefore does not seem likely from these results that individuals prioritize the changes they intend to make, i.e., they do not merely want to change the circumstances in the area with which they are least satis®ed. It seems rather that individuals
6
The relative eect of a variable is the coecient of that variable multiplied by the variance of that variable. Using the variances given in Table 2, we can see that this is highest for age.
7An interesting sideline is that when individuals are asked whether they intend to change anything,
intend changing any sphere with a low satisfaction value. The eects of the absolute value of the satisfaction variables are always quite signi®cant.
In Table 3, the results are given for the probability that German respon-dents intend changing their housing conditions:
Again, the absolute level of the partial satisfaction involved is an impor-tant variable with the expected sign.
The eect of individual circumstances is again reasonable: the higher the education, the more likely an individual is to intend to change and the higher the age (from the age of about 11 upwards), the lower the probability of intent to change, just as in Russia. The eects of satisfaction variables are not only of the anticipated sign, but are also stronger than in the Russian case. Hence, the results for Germany are encouraging.
2.3. Actual changes
Here we look at whether individuals in Russia have actually made changes from period 1 to period 2 for the areas of work and of housing.
In Table 4, a real change in an individual's housing conditions is de®ned to have occurred when the individual moved to another house, bought or rented the current house, repaired his own house, had a lodger move in, or made another change speci®ed by the respondent. A real change in an individual's
Table 3
Intentions to change the housing situation by personal action in Germany
Housing situation Job situation
Probit t-value Probit t-value
Circumstances
ln (income) )0.02 0.5 0.001 0.1
Education 0.07 9.1 0.30 2.4
Male 0.02 0.3 0.04 0.9
ln (age) 4.22 4.0 )9.50 6.5
ln2(age)
)0.68 4.2 1.24 6.0
Satisfactions
Life satisfaction )0.02 0.9 )0.11 4.4
Satisfaction with this sphere )0.12 3.8 )0.31 9.2
Maximum satisfaction? )0.10 1.8 0.07 1.0
Minimum satisfaction )0.06 0.8 0.15 2.1
ÿ2 Log L 4147.7 3204.3
#yiequal to 1 988 504
job conditions occurred when the individual went to work for another or-ganization or started his own business.
The results for Russia in Table 4 mainly reveal a tale of insigni®cance. Only gender, age and job satisfaction are signi®cant explanatory variables for whether a change occurred in the job situation. As expected, males are more likely to change their job and the likelihood decreases ever more sharply after 18 years of age. The fact that the higher the satisfaction with the job the previous year, the less likely a change is, is consistent with our expectation, but the explanatory power is weak, probably due to the high volatility of the situation in the Russian republic over the 1993±1994 period: in the present uncertain Russian situation events which are uncontrollable and unforeseen by respondents may be the main determinants of changes. We also note that the number of actual changes is less than the number of intended changes.
There were no signi®cant factors in the explanation of changes in housing conditions. There simply seem to be too many random changes between years for any variable to have a signi®cant in¯uence on the housing changes. In-deed, we can see from both housing and job changes that the actual per-centage of individuals making a change is about a third of the perper-centage of individuals who had an intention to change a year back, which is another indicator of the volatility of actual changes.
One encouraging ®nding is that there was no residual eect from the in-tention to change: when the inin-tention to change the last period was included
Table 4
Choices in housing and work in Russia from 1993 to 1994 (Probit)
Housing sit. t-value Job sit. t-value
Circumstances in 1993
ln (income) 0.13 1.7 )0.02 0.2
Education 0.02 1.8 )0.01 0.8
Male )0.02 0.3 0.53 4.2
Life satisfaction 0.01 0.3 0.06 0.8
Satisfaction with this sphere 0.08 1.1 )0.21 2.1
Maximum satisfaction? 0.07 0.6 0.12 0.7
Minimum satisfaction 0.04 0.4 0.11 0.6
ÿ2 Log L 1374.8 536.2
Number ofyiequal to 1 272 84
in the explanation of actual choices, the eect turned out to be very small and insigni®cant, which is why the speci®cation shown does not include the in-tention to change. When dropping all explanatory variables and including only the intention to change, the intention to change was signi®cant. This means that, insofar as there is any predictive power in the intention to change, this predictive power is fully encompassed by the variables used to analyze actual changes. Hence it seems that although we for instance do not have personality traits as explanatory variables, there simply are no per-sonality traits which, apart from their eect via the variables shown, have an eect on the intention to change that is predictive to actual changes. Therefore, insofar as there are variables that predict both intentions and changes, we seem to be using them.
There has been one previous study I am aware of that has used satisfaction variables to predict choices, which is the study of Clark et al. (1998) who looked at the eect of job satisfaction on future quit-behavior using 10 years of the GSOEP. They found that a lower job satisfaction level signi®cantly increased the chance of quitting in the future, though the eect found was small. No intentions, general satisfaction or relative level of job satisfaction were used.
For Germany, Table 5 shows more signi®cant results:
Table 5
Real changes from 1993 to 1994 in jobs and family life in Germanya
Divorced/ separated?
t-value Quit job? t-value
Circumstances in 1993
Life satisfaction )0.10 1.6 0.03 0.5
Satisfaction with this spherea
)0.20 2.4 )0.15 2.2
Maximum satisfaction? 0.19 1.2 0.12 1.0
Minimum satisfaction )0.53 2.8 )0.06 0.3
ÿ2 Log L 495.3 868
581 122
N 4119 2328
a
We ®nd again that the more dissatis®ed an individual was with the area in question the previous year, the more likely it was that an individual would choose to change his circumstances. Though the eects found are not very signi®cant, it was the case that the intention to change housing had no residual in¯uence on actual changes in housing. We therefore ®nd again that we seem to be using all the variables that predict both intentions and changes.
The eect of individual circumstances is similar to that in Russia.
On the basis of the results for Germany and Russia it seems that satis-faction variables have some, but limited, explanatory power for choices and intentions. There are so many changes from one year to the next and there are so many circumstances beyond the control of the individual (opportu-nities for promotion, the attitudes of other household members, luck, etc.), that the relationships between the explanatory variables and actual changes are weak. This holds especially for the data set of the Russian Republic and less so for the West-German data set.
Nevertheless, the signi®cant relationships are as expected: the lower the partial satisfaction with an area in the last period, the higher the probability of a change in this period.
2.4. Importance factors and weights
Now we look at whether individuals ®nd the areas of their lives with which they are satis®ed, more important. First, we analyze the question posed in the German panel data set of whether success in the job was important to the individual (many other importance questions were posed, but only this one has a clear corresponding partial satisfaction question to accompany it). Keeping in mind that the higher the dependent variable, the more important an individual thinks that success in his job is to him, we ®nd the results presented in Table 6.
Here the results rather conform to expectation, as the importance of work increases with the satisfaction with work. The coecients of the dummy variables denoting the relative place of satisfaction with work are also as expected.
higher level of ambition. We may hence con®rm the hypothesis that satis-faction with work is positively correlated with the subjective importance of work.
The second way we investigated whether individuals adapt their opinions as to how important several areas of their lives are for their general satis-faction, was by postulating that if an individual was very satis®ed with a sub®eld in the previous period, this sub®eld is going to be more important to the individual in this period. If for instance an individual is very unsatis®ed with his housing conditions, he may adapt to the situation by spending less and less time in his house and may try to spend less time thinking about his housing conditions. The mechanism of reducing the importance of those areas of one's life with which one is very dissatis®ed may hold for all partial satisfactions. We thus predict that the importance of an area of one's life for current life satisfaction depends on the partial satisfactions of the previous period (PEj;tÿ1):
GElife;t X0b
XM
j1
ajcjPEj;tÿ1PEj;tdGElife;tÿ1it:
The right-hand side of this equation has three structural components. The ®rst component, X0b, denotes the eect of exogenous variables on general satisfaction. The second component, PMj1 ajcjPEj;tÿ1PEj;t; denotes the
Table 6
The importance of job succes in Germanya
Importance of job t-value
eect from partial satisfactions on general satisfaction in which one can in-terpret the term ajcjPEj;tÿ1 as the weight of the current satisfaction level on general satisfaction. As one can see, the higher cjPEj;tÿ1, the higher the eect of the current partial satisfaction level on general satisfaction. Hence,cj denotes the in¯uence of the partial satisfaction in the previous period on the weights of the partial satisfaction in this period, and we expect this coecient to be positive.8The third component,dGElife;tÿ1;which is present in only one of the two speci®cations shown, is used to capture the eect of variables which we do not have available in the data but which are correlated over time.
We tried a least-absolute-deviations (LAD) estimator with bootstrapped con®dence intervals in order to see whether the assumption of normality of the error-terms had any eect in this case. Because the dierences between simple OLS and this semi-parametric technique were minute, the assumption of normally distributed error-terms is reasonable in this case and we only show the OLS results here.
The results for Germany and Russia are summarized in Table 7.
The results tell a mixed tale: when the general satisfaction of the previous period is not used as a control, all the c's are positive and their combined effectis signi®cantly positive. When the general satisfaction of the previous period is used as a control, virtually all the c's are positive, but most are insigni®cant. It hence seems likely that part of the effect of the c's in the speci®cation without the general satisfaction of the previous period pick up part of the effect of the omitted variable GElife;tÿ1. As most of the c's are positive in both tables though, there is some indication of a selective memory effect.
It may be the case that adaptation takes less than one year, the time be-tween the available panel waves. An individual experiencing a very low sat-isfaction level in one area of his life may start ignoring this area of his life in a matter of weeks rather than years. Indeed, the study by (Suh, Diener & Fujita (1996)) suggests that only the events of the last three months are relevant for self-reported satisfaction levels. If this is also the case in the yearly panel data we use here, we would expect more signi®cant results with more frequently interviewed respondents. Of course, this is mere guesswork at this point.
8The results for this reduced-form model are only used for this hypothesis. For a much more in-depth
We thus ®nd only limited evidence that respondents claim to ®nd an area more important if they were more satis®ed with that area in the past.
3. Conclusions
We set out trying to see whether an individual's verbal and choice behavior was consistent with the hypothesis that they are trying to maximize self-re-ported levels of satisfaction.
The respondents we investigated were more on average likely to intend to change those areas they are unsatis®ed with in this period, were more likely to actually have changed those areas they were unsatis®ed with in the last
Table 7
The weights of satisfactions in previous years on general satisfactiona
Russia 1994 Germany 1993
Health sat a 0.19 9.6 0.21 10.4
c )0.006 2.3 0.001 0.4
Family sat a 0.18 5.3 0.19 5.4
c )0.001 0.5 0.006 0.1 Social sat a 0.17 4.8 0.17 4.7
c 0.002 0.7 0.005 1.1
Leisure sat a 0.08 4.6 0.09 5.2
c 0.007 2.8 0.01 3.7
Gen sattÿ1 0.14 3.8 0.26 12.4
X0b
ln (income) )0.05 0.7 )0.05 0.7 0.02 0.8 0.02 0.9
Education 0.01 1.5 0.01 1.6 0.003 0.6 0.002 0.4
Male )0.01 0.4 )0.02 0.3 )0.03 1.2 )0.03 1.0
period, and tended to ®nd the areas of their lives they are dissatis®ed with less important. The eects were not strong however, though more signi®cant for the German GSOEP data set than for the Russian data set. The combined results give only limited support to the hypothesis that individuals indeed try to maximize self-reported levels of satisfaction. At the very least, self-re-ported levels of satisfactions were signi®cant determinants of intentions and choices, alongside `objective' variables such as age, education, gender and income. A major drawback was the unavailability of personality factors to include in the analyses.
One of the implications for future research is that year-by-year observa-tions on individuals may be spaced too far in time to capture the importance of satisfaction levels on choices because circumstances change quickly and the memory of many events fades rapidly.
Acknowledgements
The author would like to thank two anonymous referees for helpful comments on earlier versions of the paper.
Appendix A
In this appendix, the variables used in the analyses are de®ned and ex-plained.
A.1. Satisfaction questions
The satisfaction questions are all translated from a 0±10 scale to an
ÿ1;1 scale by inversely relating them to the empirical distribution amongst households. It may be noted that re-scaling makes it easier to work with these variables without changing the preference structure or qualita-tively altering any of the results one would get by analyzing the raw ques-tions. Some of the satisfaction questions are dierently worded amongst the two panels, but the dierences are assumed to be small, unless mentioned as otherwise in the main text. De®nitions:
Life satisfaction in Russia: ``How satis®ed are you with your life as a whole in the present time on a scale of 1±10''.
whole in the present time on a scale of 0±10''.
Financial satisfaction (R): ``How satis®ed are you with the ®nancial situ-ation of your family in the present time on a scale of 1±10''.
Financial satisfaction (G): ``How satis®ed are you with your household income on a scale of 0±10''.
Housing satisfaction (R): ``How satis®ed are you with your current hous-ing conditions (1±10)''.
Housing satisfaction (G): ``How satis®ed are you with your current dwell-ing (0±10)''.
Job satisfaction (R): ``How satis®ed are you with your current job (1± 10)''.
Job satisfaction (G): ``How satis®ed are you with your current work (0± 10)''.
Family satisfaction (R): ``How satis®ed are you with your marriage at the current time (1±10)''.
Family satisfaction (G): ``How satis®ed are you with your leisure time at the current time (0±10)''.
Social satisfaction (R): ``How satis®ed are you with your social contacts at the current time (1-10)''.
Health satisfaction (G): ``How satis®ed are you with your health time at the current time (0±10)''.
Environment satisfaction (G): ``How satis®ed are you with your environ-ment at the current time (0±10)''.
From these raw numbers, call themsijt(the satisfaction level of individuali in sphere j in period t), we get the empirically used Sijt de®ned by the fol-lowing transformation: Sijt Nÿ1 sijt; sjt;var sijt whereby sjt denotes the empirical average of the satisfaction sphere j and var sijt denotes the em-pirical variance of the satisfaction sphere j at time t. See Frijters (1999, chapter 5) for a more detailed explanation of how this works.
A.2. Intentions
The dummy designating the intention to change the job situation (R) is de®ned as one if the respondent gave an answer indicating a speci®c in-tent to act to the question: ``How are you going to change your job life''. The dummy designating the intention to change the housing situation (G) is de®ned as one if the respondent gave a speci®c area of housing which he intended to improve in the coming years.
The dummy designating the intention to change the job situation (G) is de-®ned as one if the respondent said it was certain that he would either quit his job, change jobs or would reduce working hours in the next two years.
A.3. Importance factors, personal characteristics, ®nancial need
The importance of the job (G) was de®ned by the answer to the question: ``How important is success in your job to you: very important/impor tant/not very important/unimportant''. This question was analyzed with ordered-Pro-bit techniques. The higher the latent variable the more important the sphere.
Education is de®ned in years attained. Income equals gross household income.
ldenotes the empirical log-mean of the ®ve Income Evaluation Questions: ``how much household income would your family need before taxes each month to say that your income was: very good/good/so-so/bad/very bad''. Analyses on Russia are reported withl, given the emphasis on the analysis of lin other publications. The insigni®cance of the variable for the intentions, importances and changes is also present for Germany, where we report the analyses withoutlas a large proportion of German respondents do not have l-values. It may be noted that for the smaller sample of individuals for which we did have German l-values, it was found to be an insigni®cant variable.
The dummy variable Separated (G) is de®ned as one when the respondent indicates that (s)he has been separated from his/her partner over the last year. If the individual has quit the job or terminated the contract prematurely, the individual is regarded as having quit the job.
All intentions and actual changes are (0,1) dummy variables.
References
Bentham, J. (1798).Introduction to the principles and morals of legislation. London: Athlone Press. Clark, A., Georgellis, Y., & Sanfey, P. (1998). Job satisfaction, wage changes and quits: Evidence from
Germany.Research in Labor Economics,17.
Diener, E., Suh, E., & Oichi, S. (1997). Recent ®ndings on subjective well-being.Indian Journal of Clinical Psychology.
Duesenberry, J. S. (1949).Income, saving, and the theory of consumer behaviour. Cambridge, MA: Harvard University Press.
Festinger, L. (1957).A theory of cognitive dissonance. Evanston, IL: Row-Perterson.
Frijters, P. (1998). The sale of relational capital through tenure pro®les and tournaments. Tinbergen Discussion Paper 046/3.
Frijters, P. (1999). Explorations of welfare and well-being. Ph.D. Thesis, Amsterdam: Thela Thesis Publishers.
Gilad, B., Kaish, S., & Loeb, P. D. (1987). Cognitive dissonance and utility maximization: A general framework.Journal of Economic Behaviour and Organisation,8, 61±73.
Gollwitzer, P. (1993). Goal achievement: the role of intentions. In W. Stroebe, & M. Hewstone,European review of social psychology, Vol. 4(pp. 141±185). New York: Wiley.
Heckhausen, J., & Schultz, R. (1995). A life-span theory of control.Psychological Review,102, 284±304. Helson, H. (1964).Adaptation-level theory: An experimental and systematic approach to behaviour. New
York: Harper & Row.
Horowitz, J. L. (1992). A smoothed maximum score estimator for the binary response model.
Econometrica,60, 505±531.
Horowitz, J. L. (1993). Semiparametric estimation of a work-trip mode choice model. Journal of Econometrics,58, 49±70.
Kahneman, D. (1998), Assessments of individual well-being: A bottom-up approach. In: D. Kahneman, E. Diener, & N. Schwarz,Foundations of hedonic psychology: Scienti®c perspectives on enjoyment and suffering. New York: Russell Sage Foundation.
Kahneman, D., Diener, E., & Schwarz, N. (1998). Foundations of hedonic psychology: Scienti®c perspectives on enjoyment and suffering. New York: Russell Sage Foundation.
Kahneman, D., Fredrickson, D. L., Schreiber, D. A., & Redelmeier, D. A. (1993). When more pain is preferred to less: Adding a better end.Psychological Science,4, 48±67.
Kahneman, D., Knetsch, D., & Thaler, R. (1991). The endowment eect, loss aversion, and status quo bias.Journal of Economic Perspectives,5, 193±206.
Kahneman, D., & Schnell, L. (1992). Predicting utility. In R. Rogarth,Insights in decision making(295±310). Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision risk.Econometrica,47,
363±391.
Kapteyn, A. (1977). A theory of preference formation. Ph.D., Leyden: Leyden University. Loewenstein, G., & Adler, D. (1995). A bias in the prediction of tastes.Economic Journal, 929±937. Pettigrew, J. (1979). The ultimate attribution error: Extending Allport's cognitive analysis of prejudice.
Personality and Social Psychology Bulletin,5.
Schwarz, N. (1995). What respondents learn from questionnaires: The survey interview and the logic of conversation.International Statistical Review,63(2), 153±177.
Shizgal, P. (1998). On the neural computation of utility: Implications from studies of brain stimulation reward. In D. Kahneman, E. Diener, & N. Schwarz,Foundations of hedonic psychology: Scienti®c perspectives on enjoyment and suffering. New York: Russell Sage Foundation.
Simonson, I. (1990). The eect of purchasing quantity and timing on variety-seeking behaviour.Journal of Marketing Research,27, 150±162.
Suh, E., Diener, E., & Fujita, F. (1996). Events and subjective well-being: Only recent events matter.
Journal of Personality and Social Psychology,70(5), 1091±1102.
Thaler, R. (1980). Towards a positive theory of consumer choice.Journal of Economic Behaviour and Organisation,39, 36±90.
Tversky, A., & Fox, C. (1995). Weighing risk and uncertainty.Psychological Review,102(2), 269±283. Van de Stadt, H., Antonides, G., & Van Praag, B. M. S. (1984). Empirical testing of the expected utility
Van Praag, B. M. S. (1971). The welfare function of income in Belgium: An empirical investigation.
European Economic Review,2, 337±369.
Van Praag, B. M. S., & Frijters, P. (1999). The measurement of welfare and well-being; the Leyden approach. In D. Kahneman, E. Diener, & N. Schwarz,Well-being: Foundations of hedonic psychology
(pp. 413±433). New York: Russell Sage Foundation.
Veblen, T. (1899).The theory of the leisure class. New York: Viking Press.
Veenhoven, R. (1996). Developments in satisfaction-research.Social Indicators Research,37, 1±46. Wagner, G. G., Burkhause, R. V., & Behringer, F. (1993). The English language public use ®le of the
