Here is a simple model in Stata
Code:
xtreg workhours c.age i.sex##i.crisis, fe vce(cluster id) note: 2.sex omitted because of collinearity. Fixed-effects (within) regression Number of obs = 347,296 Group variable: id Number of groups = 151,470 R-squared: Obs per group: Within = 0.0294 min = 1 Between = 0.0005 avg = 2.3 Overall = 0.0000 max = 8 F(3,151469) = 860.46 corr(u_i, Xb) = -0.2371 Prob > F = 0.0000 (Std. err. adjusted for 151,470 clusters in id) ----------------------------------------------------------------------------------- | Robust workhours4 | Coefficient std. err. t P>|t| [95% conf. interval] ------------------+---------------------------------------------------------------- age | -.2027896 .0151701 -13.37 0.000 -.2325227 -.1730565 | sex | Male | 0 (omitted) | crisis | Post-crisis | -3.43855 .1232942 -27.89 0.000 -3.680204 -3.196896 | sex#crisis | Male#Post-crisis | -.7873883 .1692412 -4.65 0.000 -1.119098 -.4556789 | _cons | 49.76833 .5767537 86.29 0.000 48.63791 50.89876 ------------------+---------------------------------------------------------------- sigma_u | 9.8865584 sigma_e | 6.8570233 rho | .67520132 (fraction of variance due to u_i) -----------------------------------------------------------------------------------
Code:
input double(workhours age) long sex float education long crisis 50 25 2 1 1 45 26 2 1 1 52 46 2 2 1 60 47 2 2 1 42 43 2 3 1 38 25 1 1 1 40 26 1 1 1 45 25 2 3 1 38 45 2 2 1 38 46 2 2 1 40 39 1 3 0 40 40 1 3 0 40 39 1 3 0
In the model above sex (2=Male, 1=Female) and education are time fixed for each individual and hence constant over time. The crisis variable is a dummy variable (0= before 2008, 1= post 2008) that changes over time.
The interaction of interest sex##crisis tells me whether the gap in working hours has increased or decreased following the financial crisis. The findings show that the gap increases following the crisis, however, I am interested to see whether this increase is driven by a decrease in working hours for men or by an increase in working hours for women or whether it is driven by both cases. Thus, I would like to run the average marginal effects to see graphically the association between both variables. After running
Code:
margins crisis##sex
Code:
Predictive margins Number of obs = 347,296
Model VCE: Robust
Expression: Linear prediction, predict()
-------------------------------------------------------------------------------------
| Delta-method
| Margin std. err. z P>|z| [95% conf. interval]
--------------------+----------------------------------------------------------------
sex |
Female | . (not estimable)
Male | . (not estimable)
|
crisis |
Pre-crisis | 41.91534 .0647819 647.02 0.000 41.78837 42.04231
Post-crisis | 38.05804 .0213521 1782.40 0.000 38.01619 38.09989
|
sex#crisis |
Female#Pre-crisis | . (not estimable)
Female#Post-crisis | . (not estimable)
Male#Pre-crisis | . (not estimable)
Male#Post-crisis | . (not estimable)
-------------------------------------------------------------------------------------There are two different threads on Stata list that recommends not to include the main effect. This means that my interaction should be used in the following way:
Code:
i.sex#i.crisis
If I follow that suggestion I still do not solve my issue (see below). I tried to rescale my variables, but that did not resolve the issue as well. Does anyone have any idea why this issue is arising?
Code:
xtreg workhours4 c.age i.sex#i.crisis, fe vce(cluster id)
note: 2.sex#2.crisis omitted because of collinearity.
Fixed-effects (within) regression Number of obs = 347,296
Group variable: id Number of groups = 151,470
R-squared: Obs per group:
Within = 0.0294 min = 1
Between = 0.0384 avg = 2.3
Overall = 0.0366 max = 8
F(3,151469) = 860.46
corr(u_i, Xb) = -0.1009 Prob > F = 0.0000
(Std. err. adjusted for 151,470 clusters in id)
-------------------------------------------------------------------------------------
| Robust
workhours4 | Coefficient std. err. t P>|t| [95% conf. interval]
--------------------+----------------------------------------------------------------
age | -.2027896 .0151701 -13.37 0.000 -.2325227 -.1730565
|
sex#crisis3 |
Female#Post-crisis | -3.43855 .1232942 -27.89 0.000 -3.680204 -3.196896
Male#Pre-crisis | 4.225939 .1182001 35.75 0.000 3.994269 4.457608
Male#Post-crisis | 0 (omitted)
|
_cons | 47.52087 .5846898 81.28 0.000 46.37489 48.66685
--------------------+----------------------------------------------------------------
sigma_u | 9.4494496
sigma_e | 6.8570233
rho | .65506221 (fraction of variance due to u_i)
-------------------------------------------------------------------------------------
0 Response to Why margins command is returning "not estimable" using Fixed effects methods?
Post a Comment