MEMO120.WP5

Re:   Choice of job importance items to be included in telephone
      and mail questionnaires
Date:       7/22/92

During pretesting the telephone instrument had incorporated 10
items measuring the importance of various aspects of jobs,
compared to the importance of high pay.  These items were based
on some used by Jencks, Rainwater and Perman (1988.  "What Is a
Good Job?  A New Measure of Labor-Market Success."  American
Journal of Sociology 93 (6): 1322-135). 

We selected 4 of these items to be retained on the telephone
instrument.  The original 10, including these 4, will now become
part of the mailed questionnaire.

For convenient reference, here are the 10 items for the mail
instrument:

23.   Please compare the importance of each of the following job
      characteristics with the IMPORTANCE OF HIGH PAY.

            Circle the number that best describes the IMPORTANCE 
            of each characteristic COMPARED TO HIGH PAY

            More Important          Same       Less Important

300m  Having the opportunity to get on-the-job training. 
      
302   Having a large number of paid vacation days.

302f  Being able to do different things rather than the same
      things over and over.

302m  Having a low risk of losing your job.

302s  Being able to decide what time to come to work and when to
      leave.

304   Being able to work without frequent checking by a
      supervisor.

304f  Being able to avoid getting dirty on the job.

304m  Having a job that other people regard highly.

304o  Having a job that provides health insurance. 

304q  Having a job that provides a pension plan.

We used the Jencks et al. results and a quick analysis of the
second-to-last pretest of about 28 non-sample respondents to help
choose which items would remain on the telephone instrument.  We
were looking for the strongest items which might have independent
effects on respondents' perceptions of job quality.  

Finally, here are numbers for 7 of the 10 items as reported by
Jencks et al. from the regression of job ratings on job
characteristics (op cit., Table 2, p. 1336):

Item     Jencks      Content

300m      .134       Having the opportunity to get on-the-job
                     training. 
         
302       .107       Having a large number of paid vacation days.

302f     -.096       Being able to do different things rather than
                     the same things over and over.

302m     -.073       Having a low risk of losing your job.

302s      .106       Being able to decide what time to come to work
                     and when to leave.

304      -.104       Being able to work without frequent checking by
                     a supervisor.

304f      .132       Being able to avoid getting dirty on the job.

304m     (n.s.)      Having a job that other people regard highly.

304o     (n.a.)      Having a job that provides health insurance. 

304q     (n.a.)      Having a job that provides a pension plan.

Note that Jencks's regression is on measures of the presence of
these characteristics in R's current job, rather than on measures
of the expressed importance of these items to R.  Jencks reported
that status (304m) was not significant.  Having a pension and
health insurance may or may not have been in the list of items he
screened and rejected as nonsignificant.Conclusions
  
Based on the above, and on spinning the 3-D graphic of the
principal components and items (using the JMP Macintosh program),
we made the following selections:

Autonomy item.  Items 302f (do different things), 304
(supervision frequency) and 302s (decide own hours) formed a
cluster in the principle components space.  Because its spread
was better than 302s's, because it had a good _ of -.104, and
because it had good face validity as a measure of autonomy, we
chose item
304 (supervision frequency) for retention in the phone
instrument.

Risk aversion item.  Items 302m (low risk of losing job), 304o
(health insurance), and 304q (pension plan) formed a cluster. 
304o had the worst univariate distribution, with other 2 roughly
comparable.  Because 304q (pension) seemed more clearly relevant
to the topic of our questionnaire, we chose it to remain on the
phone instrument.

On-the-job training item.  The remaining 4 items did not form
obvious clusters.  Since 300m (on-the-job training) had a high _
of .134, and a good univariate distribution, it was selected for
retention.

Status of job item.  The 3 items then remaining did not obviously
cluster either statistically or substantively:  304m (job highly
regarded); 302 (number of vacation days); and 304f (get dirty). 
304f (dirty) had a high _ of .132, while 302 (vacation) had a
respectable .107 and the status measure corresponding to 304m
(regard) was reported by Jencks to be insignificant.  However,
the univariate distribution of 304f was poor, while the other 2
were good.  Because of our interest in status, and despite the
strong location of 302 (vacation) in the principal components
space, we chose to keep 304m (regard).


We are under no particular illusions about the strength of our
evidence, based on 25 cases, but the above comments describe the
procedure we used nonetheless.