I’m running a regression in which the outcome is the change in log hourly wages between two time periods (l_occ_hwage_chg). To generate the outcome variable, I took the difference between log hourly wages in month t+1 with log hourly wages in month t. I want to understand a) how the outcome (change in log hourly wages) changes over two 6-month time periods (timeperiod==1 (months 7-12); timeperiod==0 (months 1-6)); and b) how the change over the two time periods differs between two sets of states: one where a policy of interest is in place (policy==1); and another where the policy is NOT in place (policy==0).
I’m using the following code; and follow it with a margins command that--I think--tells me how much the outcome (l_occ_wage_chg) changes between timeperiod==1 and timeperiod==0, in states where policy==1; and in states where policy==0. My question: how do I interpret the output of the marginal effects command? Pasting some sample output below. My outcome is the change in two logged values. So, looking at the pasted output, is it accurate to exponentiate -.056406 and -.1061071? How do I interpret the difference in these values?
I’m using the following code; and follow it with a margins command that--I think--tells me how much the outcome (l_occ_wage_chg) changes between timeperiod==1 and timeperiod==0, in states where policy==1; and in states where policy==0. My question: how do I interpret the output of the marginal effects command? Pasting some sample output below. My outcome is the change in two logged values. So, looking at the pasted output, is it accurate to exponentiate -.056406 and -.1061071? How do I interpret the difference in these values?
Code:
regress l_occ_wage_chg policy##timeperiod `controls’ … margins i.policy, dydx(i.timeperiod) pwcompare(cimargins effects)
Code:
margins i.policy, dydx(i.timeperiod) pwcompare(cimargins effects)
Pairwise comparisons of average marginal effects
Model VCE: Robust Number of obs = 7,118
Expression: Linear prediction, predict()
dy/dx wrt: 1.prepostperiod_ma
---------------------------------------------------------------------
| Delta-method Unadjusted
| Margin std. err. [95% conf. interval]
--------------------+------------------------------------------------
0.timeperiod | (base outcome)
--------------------+------------------------------------------------
1.timeperiod |
policy |
0 | -.056406 .0573843 -.1716658 .0588539
1 | -.1061071 .0592991 -.2252127 .0129986
---------------------------------------------------------------------
Note: dy/dx for factor levels is the discrete change from the base
level.
-------------------------------------------------------------------------------------
| Contrast Delta-method Unadjusted Unadjusted
| dy/dx std. err. t P>|t| [95% conf. interval]
--------------------+----------------------------------------------------------------
0.timeperiod | (base outcome)
--------------------+----------------------------------------------------------------
1.timeperiod |
policy |
1 vs 0 | -.0497011 .0233979 -2.12 0.039 -.0966971 -.0027051
-------------------------------------------------------------------------------------
Note: dy/dx for factor levels is the discrete change from the base level.
Comment