Following on from the second workshop a month ago, today saw the third in the series of “Workshops on Well-being” take place at the LSE. This time the presentation was given by Andrew Clark of PSE. Below are some (very) impressionistic notes.
Presentation by Andrew Clark on Job Satisfaction: What do we Know and What Next?
- Job satisfaction (JS) and individual well-being (LS)
- well-being/LS function LS = f(Job satisfaction, health satisfaction, leisure etc)
- data in BHPS (waves 6-15 though 11 missing)
- health/ income / house / spouse / job / social life / amout leisure / use leisure (scale 1-7)
- all highly significant
- social life is top, followed by health, use of leisure, income and job satisfaction is last
- robust to demographic controls
- But do individual personality types make any difference (fixed effects)
- Panel results
- all effects go down (there are ‘happy types’) except JS (which doubles) though still the smallest
- Is this ranking unique to Britain?
- Is it the same for everyone? (subregressions: old/young, men/women; or do a latent class analysis)
- JS is important to firms as well as it will predict worker behaviour
- Labour turnover
- Absenteeism
- Counter-and non-productive work/productivity
- Worker quitting (but almost impossible to do properly as quitting is self-reported so unreliable)
- P(quit(t+1)) = f(JS, X(t))
- Compare quitting GB and Germany
- pretty similar, JS is pretty significant
- Cognitive biases and context in relation to quitting (SPELL data from BHPS)
- have panel data so can look at series of JS for an individual
- refers to Kahnemann and Riedelmeier on evaluation of colonoscopy
- suggest Peak-End evaluation: evaluations of peak and end point
- Apply to job quitting (peak-end, min, max, avg, current …)
- peak-end does best (followed by running max (close), and current)
- => behaviour would not then seem to max their utility
- Try do the same with income but need variations in income (since normally just rises)
- use truckers as they have exog changes
- other potential sources: tax changes
- peak-end divorce
- Relative income
- Traditional: W/LS/JS = W(y,…)
- Comparisons: LS/JS = W(y/yr, …)
- yr is comparison/relative income
- to whom do we compare? (peers, others in HH, spouse, myself in the past, friends, neighbours, work, expectations)
- Results:
- +ve effect of income
- But falling as other’s income rises
- Overall effect is zero: if everyone’s income rises then no effect
- Preference for structure of income
- same income but in different ways
- flat slope (A), steeper (+ve) slope (B), and v. +ve slope (C)
- even though flat slope (using saving to mimic C) would result in being overall better off
- asked about this (they were told they could this) they still chose C (apparently because of self-control issues – they wouldn’t be able to save)
- Does other’s income always affect one negatively
- Hirschmann’s tunnel effect (happy for something good for you because it means something good is going to happen to me)
- Danish ECHP (1994-2001): fantastic data (which gave not only individuals but all of their colleage’s info including pay)
- here one does find a +ve effect of others income on me (check how it varies across firm so not just selection effect at firm level)
- Do 2 wrongs make a right?
- Peak-end utility could be thought of as ‘correct’ as:
- with adaptation
- current utility (after something good) understates actual total flow benefits (as one has adapted)
- PE corrects for this
- Instrumental uses of JS
- ‘Good job’ lit has mainly focused on money
- But self-employed earn less but are happier (though significat issues about reporting bias)
- Also why are there different avg. wages in different industries (when they look the same)
- Compensating differentials vs. rents
- So let’s use JS to explain different
- looking at the data: high wage goes with high JS (so suggests this about rents not compensation)
- Job Quality: Are things going to the dogs?
- ISSP (repeated XS in 3 waves 1989 - 2005)
- Multivariate regressions: JS is improving (went down 1989-1997 but bounced back in 2005)
- But stressful/dangerous/difficult work has been rising
- Good job content has been going down.
- However enough other stuff has been getting better faster (income, opportunities, flexible hours)
Discussion
- Paul Dolan:
- Causality
- Experienced Utility? Kahnemann would be unhappy
- Peak-end seems difficult for JS since already a retro-spective evaluation (so peak-end of a peak-end)
- Gordan:
- relative ranked position not just compared to the avg
- care more about those above than those below
- need to be more specific about form of relativities
- All: Context, Context, Context
- RP: Peak-end vs. range-frequency. Take colonoscopy: PE predicts that increasing pain at a single point (early on) would worsen evaluation while range-frequence would predict it would improve evaluation (since you spend more time at a level relatively better than the worst)
- BHPS: now have a question asking for whether your LS is better/worse than last year
- Gordan: gratitude is single biggest predictor of happiness
- individual differences
- Propensities to adapt
- Gordon: Andrew Oswald and he also found +ve avg income effects in workplace
- Judgment vs. Adaptation
- Paul Dolan: generally we overestimate our +ve attributes but underestimate (their relative) income level