Document Sample
					   Job Quality and Effort

Andrew E. Clark (Paris School of
      Economics and IZA)

          APE/ETE Masters Course

“Have jobs been getting worse?”


“Has job quality declined since the (mythical) golden
   age of the 1960s and 1970s?”

Nostalgia is a wonderful thing. But it is our duty to look
   at the facts, and then try to bring economic analysis
   to bear on them.

So what has happened?
1) There are now more jobs (or at least up until
Unemployment rates have mostly fallen in
    OECD countries.
Something of an “Anglo-Saxon” phenomenon.

2) The characteristics of these jobs would
    broadly seem to be better than in the past.
Which characteristics?
a) Wages have increased in almost all countries.
   One major exception is the US. Over the
   1985-’95 period, real labour income in the
   first decile fell. But so did real labour income
   in the fifth decile (and the median I think).
b) There was rising earnings inequality (see
   handout). This will reduce utility at a given
   level of mean income.
c) Hours of work are trending inexorably
d) But have jobs become less secure? Five-year
    retention rates fell sharply 1980-’95 in
    Finland, France and Spain. No strong
    movement elsewhere.
i) RR is not the only important characteristic,
    the consequences of job loss need to be taken
    into account (chances of finding another job,
    unemployment benefits).
ii) Movement between jobs might allow better
Subjective evidence on job security from three waves of the ISSP
Overall, good news might outweigh the bad.

Unfortunately, work on the time series of job
    satisfaction – workers’ evaluations of their own
    jobs – dropped sharply from the 1980s and 1990s
    into the 2000s (see the handout from Francis
    Green’s book). An exception is the US.

What’s gone wrong? One idea is what individuals
     actually do when they are at work: “job content”.
     This captures how hard they work, danger, interest
I will mostly concentrate on worker effort.
There is a small literature on accidents at work.
   Workplace accidents are found to be

i)   Higher (a little) for temporary rather than
     permanent workers.
ii) Unrelated to hours of work.
iii) Lower in unionised workplaces.

There is also a more aggregate/macro literature
   that has looked at time series movements in
   accidents – see Askenazy’s book.
The health-related consequences of work have
   worsened in Europe between 1990 and 2000
   (see handout).

The US was on the same trajectory until the early
    1990s; since 1990 the number of accidents at
    work-related illnesses have dropped by 1/3.
Why have the French and American experiences
    been so different in recent years?
1) Americans take worker health seriously
    (Ergonomics and training have long-run
    productivity payoffs).
2) Government and unions take an aggressive stance on
     workplace safety. Information on safety violations
     made public. So workers won’t work there, or will
     ask for higher wages, and insurance premia
     (private) rise.
The latter rose from 1.4% of labour costs in 1985 to
     2.4% in 1994. Dropped back to 1.6% in 2001.
In France the number of Inspecteurs de Travail has
     fallen. The results of investigations are not made
     public. There is thus less incentive to make
     workplaces safer (insurance is mutual, so we have
     the problem of the commons).
Worker Effort
We tend to write production functions as
We should probably write Q=Q(Nh,K), or better
    Q=Q(N,h,K), as workers and hours aren’t
    perfect substitutes.
Even better, let’s write Q=Q(N,h,e,K), where e
    shows the level of effort furnished by
    workers per hour of work. Firm’s profit rises
    with e; worker utility falls with e.
Effort is not contractable: we are in the world of
Could falling job quality be caused by greater worker
One way of looking at this is to trace out movements in
     overall job satisfaction, and then decompose them.
See regression in handout, using BHPS data from 1992-
     2002. This shows two regressions:
Pooled: each observation treated as if it represented a
     different person; presents a snapshot of average job
     quality in each year.
Panel: Follows the same individual from one year to
     another; picks out within subject changes in job
These regressions include “standard” controls:
    age, sex, education, marital status etc.
They also include a full set of year dummies
    (1992 is the omitted category). These plot the
    conditional movements in overall job
This falls pretty much monotonically, both in
    pooled and in panel regressions.
            Table 6. Overall Job Satisfaction Regressions. BHPS 1992-2002.

                                      Pooled          Panel
1993                                  -0.079*       -0.165**
                                      (0.033)        (0.042)
1994                                 -0.127**       -0.241**
                                      (0.033)        (0.054)
1995                                 -0.142**       -0.256**
                                      (0.033)        (0.069)
1996                                 -0.116**       -0.231**
                                      (0.032)        (0.085)
1997                                  -0.067*         -0.195
                                      (0.032)        (0.103)
1998                                 -0.171**        -0.278*
                                      (0.032)        (0.121)
1999                                 -0.204**        -0.355*
                                      (0.031)        (0.139)
2000                                 -0.206**        -0.333*
                                      (0.031)        (0.157)
2001                                 -0.169**         -0.307
                                      (0.031)        (0.178)
2002                                 -0.200**         -0.362
                                      (0.032)        (0.197)
Male                                 -0.208**
Age                                  -0.061**         -0.009
                                      (0.003)        (0.021)
Age-Squared/100                      0.079**          0.014
                                      (0.004)        (0.010)
High Education                       -0.208**
Medium Education                     -0.153**
Separated                              0.005         0.121*
                                      (0.031)        (0.047)
Divorced                              -0.040*         0.028
                                      (0.017)        (0.039)
Widowed                                0.056          0.099
                                      (0.040)        (0.103)
Single                               -0.116**       -0.094**
                                      (0.014)        (0.033)
                                                 Overall Job Satisfaction

Regression Coefficient

                                  1992   1993   1994 1995   1996   1997   1998     1999 2000   2001   2002

                                                              Pooled       Panel
                                                       Satisfaction: Pay

Regression Coefficient

                         -0.250   1992   1993   1994   1995   1996   1997 1998      1999   2000   2001   2002

                                                                Pooled      Panel
                                                Satisfaction: Hours of Work

Regression Coefficient





                                  1992   1993   1994   1995   1996   1997   1998     1999   2000   2001   2002

                                                                Pooled       Panel
                                                 Satisfaction: Job Security

Regression Coefficient

                         -0.250   1992   1993   1994   1995   1996   1997   1998     1999   2000   2001   2002

                                                                Pooled       Panel
                                                 Satisfaction: Work Itself

Regression Coefficient

                                  1992   1993   1994   1995   1996   1997   1998     1999   2000   2001   2002

                                                                Pooled       Panel
Worse job content from greater effort?

In an efficiency-wage framework, effort rises
  due to:
- Higher wages (but higher wages raise utility)
- Higher unemployment (but endogenous….)
- Falling cost of monitoring
- Falling cost of firing shirking workers
- Greater cost of shirking for workers

I concentrate on the last two.
 Employment protection and effort
Consider absenteeism as an indicator of employee
You can be absent because you’re sick, or because you
  shirk (“pulling a sickie”).
 Most popular sick days are Monday and Friday
 Sick days correlated with holidays and sporting
 Public-sector sick rates are 44% higher than in the
  private sector (selection of worse health to public
Effect of a probationary period before permanent job:
  Ichino and Riphahn (2005).
There are three states (exogenous)

Sick (S)                             
Lazy (L)                             
Healthy and willing to work          1--

The worker decides whether to be absent or not (A=1 or A=0).

The payoff is the continuation value (asset value) of the job minus the disutility of work. This
disutility depends on whether the worker is sick or lazy).

(A=0 | U=L)          =       W – VL                               (1)
(A=0 | U=S)          =       W – VS                               (2)
Firms can decide to check on the worker’s health status.
If check and U=L then the worker is fired and the payoff, , is zero.
If don’t check and U=L then the payoff is W.
P(check) = q

(A=1 | U=L)          =      W(1-q)                               (3)
(A=1 | U=S)          =      W                                    (4)
The subtlety here is that q won’t be set high enough
to drive L down to zero, as the firm wants to identify shirkers
(before the end of the probationary period).

Analogously, P(A) when sick, comparing (2) and (4) is S =1.

The overall probability of absence is thus
 = (1 – FL(Wq)) + 
The number of monitored absences is qK, so that the number of identified lazy workers is
B = qN  (1 – FL(Wq)) = qN  L(q)

The firm wants to maximise B by choosing q:

                          q d L
dB/dq = 0 implies that            = 1, which defines optimal probationary monitoring, q*.
                          L dq

After probation, there is no monitoring (q=0). If the probationary period is called period 0, we
thus have the final (obvious) result that K0 < K1.
Does Monitoring Work?
       Nagin et al., AER (2002).
Employees are “rational cheaters”, and shirk
  more as monitoring falls.
Experimental approach. Call-centre operators at
  16 sites, soliciting donations. Piece-rate: pay
  rises with no. donations.
Pledges can’t be linked to individual callers: no.
  of succesful calls self-reported by operators.
Monitoring: re-ring individual to check if they
  pledged (callback). This is expensive.
Operators who “cheat” have pay docked, and
  may be fired.
Manipulate the call-back rate experimentally in four of
 the 16 sites.

As EW theory would predict, the number of “bad calls”
  responds to the call-back rate.

Heterogeneity in worker response. Those with “positive
  attitudes” respond less to monitoring.

And attitudes are shown to be function of y*, estimated
 from a wage equation: the more others like me earn,
 the less positive are my attitudes, and the more
 responsive I am to opportunities to cheat.
 McVicar, Labour Economics (2008).
Considers job-search effort by the unemployed,
  rather than work effort by the employed.
Random variation due to the refurbishment of
  Benefit Offices in Northern Ireland.
Subsequent periods of zero monitoring of the
  unemployed were associated with a 16% fall in
  all exits from unemployment.
This effect particularly strong for exits to
  employment: consistent with lower job-search
 Temporary Jobs and Work Effort.
Temporary employment is on the rise: stepping stones to good
So temporary workers have a greater incentive to supply effort
     (the rewards are greater).
Engellandt and Riphahn, Labour Economics, 2005.
Swiss LFS data.
Effort measured by absenteeism and unpaid OT.
Observe that P(Temp  Perm) positively correlated with worker
     effort when Temporary. Workers assumed to prefer
     permanent to temporary jobs.

                     Perm                 Temp
Absence              1.2                  0.8
UOT                  20.6                 27.7
Extensions (to standard EW)

What about the workers, who have been pretty mute so far?
Think of a potential role for unions: effort might be bargained
Clark and Tomlinson (2001).
Data from Employment in Britain, 1992.
Measure discretionary effort:
“How much effort do you put into your job, beyond what is
Immodest replies (N=2700):

     Effort    %
     None      3
     Little    6
     Some      23
     Lot       68
                    Regression for Effort
Econometrics shows that effort rises with:
a) Wage
b) Liking hard work (slope of IC)
c) Ease of dismissal
d) Performance pay

Effort falls with
f) Male
g) Unions
These are multivariate results, so the union effect is conditional
  on wages.
                The Psychology of Effort
Any role for income comparisons: e = e(y/y*)?
I feel hard done by (relatively) by my firm, so I provide less effort.
Clark, Masclet and Villeval (2010)
Survey data from the 1997 ISSP on discretionary effort;
and a gift-exchange game in the laboratory.
Main Results:
1) Field and Experimental produce the same results
2) e = e(y/y*) indeed
3) Rank matters more than ratio (comparisons are ordinal)
4) The more I earned in the past, the less hard I work today for any
   given wage (habituation).
               Effort and Loss-Aversion
Abeler, J., Falk, A., Goette, L. And Huffman, D., "Reference
  points and effort provision". American Economic Review,

Experimental approach.

Subjects work on a tedious task: counting the number of zeros in
  tables that consisted of 150 randomly ordered zeros and ones.
Two stages
During the first stage, subjects had four minutes to count as many
  tables as possible. They received a piece rate of 10 cents per
  correct answer for sure.
• Count zeros in tables shown on
  the screen

   – Boring and pointless task
   – Very low intrinsic motivation
In the second (and main) stage, the task was again to count zeros, but
   there were two differences compared to the first stage.
First, they could now decide themselves how much and for how long
   they wanted to work. At most, they could work for 60 minutes.
How much subjects chose to work is the main outcome variable in
  the analysis of effort.

The second difference was that subjects did not get their accumulated
  piece rate earnings from the main stage for sure. Before they
  started counting in the main stage, they had to choose one of two
  closed envelopes. They knew that one of the envelopes contained a
  card saying “Acquired earnings” and that the other envelope
  contained a card saying “3 Euros.” But they did not know which
  card was in which envelope.
Uncertainty is resolved only after they have stopped working
There were two main treatments. The only difference between these
  treatments was
the amount of the fixed payment: 3 Euros or 7 Euros. Treatments
   were assigned randomly to subjects.
If the fixed payment is f, the piece rate is w and effort is e:

 Optimal effort e* is independent of the fixed payment, f.
This makes sense, and underlines an important economic truth: for a
  variable (price, others’ actions, whatever) to affect my behaviour,
  it must affect the net marginal utility (= marginal utility –
  marginal cost) from my actions. A deadweight effect on utility is
  like a sunk cost and won’t change behaviour.

The findings are that those in the 7 Euro fixed payment treatment
  work significantly longer before stopping.

How can this be explained?
Is this just tracing out a labour supply curve? Have we just shown
    dH/dw > 0?
No, because you receive f (with probability of 0.5) whether you work
  for 30 seconds or the full hour.
Many subjects stop when accumulated earnings equal the fixed
   payment (continuous updating of no. of tables correctly
                                             • HI vs. LO (N=120)

                                             Stopping at 3 euros
                                             • LO: 15.0 %
                                             • HI: 1.7 %
                                             • U-test: p=0.009

                                             Stopping at 7 euros
                                             • LO: 3.3 %
                                             • HI: 16.7 %
                                             • U-test: p=0.015

                                             Stopping at f modal
                                             choice in both
The authors argue that f affects H via the marginal utility of piece
  rate earnings (=we, which are received with probability of 0.5).

Often, people compare their outcome to some reference point, as in
  loss aversion (Kahneman & Tversky 1979)

 Paying an unexpectedly high price for a good
 Not getting an expected wage increase
 Being rejected after a "revise & resubmit" vs. being rejected

Specifically, f acts as a benchmark, and earning less than f (from
  the piece rate payment) is perceived as a loss. Individuals are
  loss-averse and thus act to reduce the chance that this happens.
Greater effort is therefore associated with higher
  expectations or benchmarks. In this experiment there is
  a probability that you will receive the benchmark, but
  we could imagine this in general being socially-

Higher expectations will lead to greater effort. But does
  that mean that worker utility is lower? In this
  experiment it is unclear whether worker well-being is
  lower as f rises (as they may well receive this fixed
  payment). More generally, if there is no utility value
  from the benchmark, greater expectations should
  correspond to lower utility.
Job Quality fell from the 1980s/1990s to the early 2000s. Job content
  might have been the reason. Interpretations.

1) Wages went up (but that doesn’t reduce utility)
2) Unemployment increased (but endogenous, and false)
3) The cost of monitoring fell
4) Easier to sack shirkers
5) Consequences of shirking now more serious (more tournaments)
6) Declining unionism
7) A possible psychological role for effort (but this doesn’t
   work..things that make me work hard should also make me
                          So What?
Why do we care about job quality?
Because it is a measure of VE, the value of a job. And we worry
  about this for social welfare reasons.
But also because it might help us to understand labour-market
The value of a job is relative to unemployment or inactivity.
As VE – VU falls, employment becomes less attractive. This can
  happen because job quality falls, or because unemployment
  becomes less unpleasant (for example, the social-norm effect).
BHPS Results from Clark (2003)
GSOEP Results from Clark et al. (2010)

Shared By: