Negative Marginal Effects

[Anna O'Donnell]

Hi Everyone, I'm hoping ye can put my mind to ease. I'm looking at the effect of increasing job insecurity on job satisfaction among public and private sector individuals. Naturally my coefficients are negative - as you increase in insecurity your job satisfaction decreases.

However my marginal effects for predicting an 8 on the job satisfaction scale for public and private sector individuals are as follows: (dy/dx(9))

public = -.083***
private = -.068**

Am I right in saying that increasing job insecurity reduces the probability of reporting a 9 on the job satisfaction scale by a greater amount for public sector workers. Is the adverse impact of job insecurity greater for public sector workers?

Technically -.083 is a smaller number than -.068 but I do not think these absolute terms apply to interpretations of marginal effects.


Any information you have is truly appreciated.


Kind regards,
Anna O'Donnell

[Richard Williams]

Anna it would help very much to see your commands and output. Just based on what you have said I can't even tell if you used regress, logit, ologit, or what. I also don't know exactly what margins command you gave.

[Anna O'Donnell]

I ran the user written command -cmp for a series of two equations within a simultaneous system. This command is specified so it is similar to the bioprobit command where each equations is estimated by ordered probit. The following is my regression command:



cmp (jobsatisfaction = jbscr2 gender age15_19 age30_39 age40_49 age50_59 age60_69 age70_79 age80_89 ///
hlth_fair hlth_bad hlth_vbad marit_civil marit_separated ///
marit_divorced marit_widowed marit_never marit_annulled ///
educ_lsecond educ_lowsecond educ_uppersecond educ_nonteritiary educ_postgrad ///
contract_unlimited contract_limited estab_size10to24 ///
estab_size25to99 estab_size100to499 estab_size500plus ///
wkhrs0to9 wkhrs10to19 wrkhrs30to39 wrkhrs40to49 wrkhrs50plus ///
unemp_5m union_current union_previous lnincome2) (jbscr2 = wkshr3y value_security difficultdummy ) ///
if paidwork == 1 & essround == 5 & publicdummy == 1, indicators($cmp_oprobit $cmp_oprobit) robust nolr

and the marginal effects are calculated for a probability outcome equal to 9:

mfx, force predict (pr outcome(9))

providing me with the following results for my public and private sector subsamples:


Public Sector Workers = -.083***
Private Sector Workers = -.068***

I am trying to figure out which group of individuals is more adversely impacted by increasing job insecurity. I'm interpreting these marginal effects as follows:

'Job insecurity decreases the probability of reporting 9 on the job satisfaction scale. This effect is greater for public sector workers'


If you have any information I would greatly appreciate it. The negative signs are making me rethink how I am interpreting the magnitude of these marginal effects coefficients.




Kind regards,
Anna