Thursday, September 09, 2010

Unemployment and Well-Being

I am currently reading through detailed transcripts of 13 focus group interviews with 100 or so people claiming unemployment benefits interviewed by Michael Egan and others as part of our IRCHSS-funded wellbeing study. I am not going to go in detail through the results as we are coding them up (along with Nicola O'Connell who works in Geary also) to present and publish them in Autumn. It is not surprising perhaps that a sense of despair hangs over many of the interviews. In general, the link between wellbeing and unemployment is one of the great puzzles of behavioural economics. The effects are, always and everywhere, enormous and poorly explained by any standard economic measures. To the very best of my knowledge, no paper has offered a convincing explanation based on financial variables alone as to why unemployment generates such substantial negative effects on well-being.

A standard OLS life satisfaction regression from the European Social Survey round 4 data collected in Ireland this year and last year confirms the basic picture.  Life Satisfaction is a simple 0 to 10 variable. The bottom two categories are unemployed and actively seeking employment, and unemployed and not actively seeking employment, respectively; Health is a one-to-five variable going from 1 excellent to 5 poor; discussint is the abscence of someone to discuss intimate problems with; gender is 1 if female and 0 if male; yearseduc-n is years of education; the base category for marital status is married and the categories are separated, divorced, widowed, never married, refused. (Results very similar for ordered logit).

      Source |       SS       df       MS              Number of obs =    1759
-------------+------------------------------           F( 23,  1735) =   10.71
       Model |  980.111862    23  42.6135592           Prob > F      =  0.0000
    Residual |  6902.76364  1735  3.97853812           R-squared     =  0.1243
-------------+------------------------------           Adj R-squared =  0.1127
       Total |   7882.8755  1758  4.48400199           Root MSE      =  1.9946

     lifesat |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
         age |  -.0802745   .1510821    -0.53   0.595    -.3765967    .2160477
      gender |  -.0723667    .100566    -0.72   0.472      -.26961    .1248766
      health |  -.5741351   .0627302    -9.15   0.000    -.6971698   -.4511004
discussint~e |  -.5104372   .1702542    -3.00   0.003    -.8443623   -.1765122
yearseduca~n |   -.019549   .0133266    -1.47   0.143    -.0456868    .0065889
_Imaritals~2 |   -.737398   .2325077    -3.17   0.002    -1.193423   -.2813732
_Imaritals~3 |  -.4806907   .3412548    -1.41   0.159    -1.150005    .1886233
_Imaritals~4 |   .4508631   .1857378     2.43   0.015     .0865694    .8151567
_Imaritals~5 |  -.4225155   .1263595    -3.34   0.001    -.6703484   -.1746826
_Imaritals~6 |   1.195028   1.426421     0.84   0.402    -1.602658    3.992713
unem~dactive |  -1.393972   .1867193    -7.47   0.000    -1.760191   -1.027754
unem~tactive |  -1.400559   .3045898    -4.60   0.000    -1.997961   -.8031574
_Ihousehol~2 |  -.0413669   .2408005    -0.17   0.864    -.5136567    .4309229
_Ihousehol~3 |  -.1167114   .2513269    -0.46   0.642    -.6096469    .3762241
_Ihousehol~4 |   .0581281   .2567047     0.23   0.821    -.4453551    .5616113
_Ihousehol~5 |   .0213614   .2638449     0.08   0.935    -.4961261     .538849
_Ihousehol~6 |  -.1041712    .273291    -0.38   0.703    -.6401856    .4318432
_Ihousehol~7 |   .1653532    .286897     0.58   0.564    -.3973471    .7280535
_Ihousehol~8 |   .1252994   .2952895     0.42   0.671    -.4538615    .7044603
_Ihousehol~9 |   .0509964   .3324416     0.15   0.878    -.6010321    .7030249
_Ihouseho~10 |   .2078842   .3228671     0.64   0.520    -.4253654    .8411338
_Ihouseho~20 |  -.0254768   .2892778    -0.09   0.930    -.5928467    .5418932
_Ihouseho~21 |   .1244626   .2963733     0.42   0.675    -.4568239     .705749
       _cons |   9.529465    .475546    20.04   0.000     8.596761    10.46217

Addendum 1: I have added household income into the specification.The base category is bottom decile. 20 and 21 are missing income flags.

This is one of the most important topics for policy during this current recession,  both in straight-forward human terms as this is one of the clearest sources of distress that emerges from a collapsed economy and also in terms of general economic welfare. The complete sense of despair and uncertainty and the shattering of confidence that appears to follow soon after redundancy for many people is a major trigger of more long-run negative consequences of recession. It is one of the main sites where economics, psychology and other disciplines need to be engaged to generate more convincing explanations with more practical solutions. It is really frustrating to think that we have advanced so far in many different ways yet our response to unemployment seems identical to that of the 1980s and is having a predictably strangling effect on people.


Kevin Denny said...

I don't know if any of the papers in this literature controlled for income, or perhaps the replacement rate, as then you'd see how much or how little income explains. But is there not also hedonic adaptation- you get used to being unemployed?

BrendanH said...

Mattias Strandh wrote a nice paper on this some years ago in Work Employment and Society, using a cohort survey of Swedish unemployed. He was able to show that financial distress and concrete expectations about the future had very clear effects on GHQ.

WES 14(3) 2000.

Enda Hargaden said...

Kevin - I'm not implying you're suggesting it's the case - but do we want that to happen?

I've always been taught that people giving up on the labour market was a bad outcome all round.

I accept it's probably the case that the coefficient will get a bit closer to zero as people give up. But for me that's something we should be looking to avoid, rather than something to console ourselves with.

It might also be interesting to look at this from a vignettes point of view. "Jim has given up on life. How well do you trust his self-reported happiness?"

Kevin Denny said...

I'm suggesting, based on the evidence that people adapt (for example to lottery wins or disabling accidents) that it will happen.
From a welfare (i.e. a normative) point it is an interesting question. After all, isn't learning to cope over adversity a good thing? Presumably yes but not if its at the expense of others which to some extent it would because of the fiscal burden of unemployment.
One could think of hedonic adaptation as a form of psychological insurance & like any insurance there are moral hazard problems.
Bottom line is: if we are doing welfare analysis then we should do it based what people's utility is not what we would like it to be.

Liam Delaney said...

I added in household income in categories Kevin, with lowest household income being the base category. I have tried other specifications and the results are very similar. In a simple linear specification using the household income categories and treating them as continuous, household income is significant. Almost none of the unemployment effect is explained.

Previous work I have done in Ireland with Carol Newman and Brian Nolan uses very detailed income specifications and accounts for unobserved heterogeneity and persistence. We were able to chip away at a small amount of the unemployment effect on financial satisfaction. In general, there are many papers controlling for income in these types of regressions and I am not aware of any that explain a sizeable amount of the wellbeing effect.

In terms of hedonic adaptation, I will add another addendum later as I have looked at the duration of unemployment, which is in the data. The usual caveat about this being a blog rather than a final draft of an AER article apply. But if I had to give an answer I would say that people do not hedonically adapt to unemployment and that, even worse, unemployment is scarring in that it leaves effects beyond its duration. There is some evidence from Clark and Oswald and others that the magnitude of unemployment effects may be related to the number of unemployed people in your social network e.g. being laid off in an area with very high unemployment may not feel as bad as being the only unemployed person in your neighborhood.

Measurement error in income and in the various income components is clearly one issue, as is the difficulty in measuring wealth and consumption.

Liam Delaney said...

The low coefficient on education is something that I have found in most of the work in Ireland. One possible explanation is the raising of standards that occurs with education. The large coefficients on health are as pervasive as the large coefficients on unemployment though people may use health and satisfaction report scales in similar ways so more thought is needed on that.