Hi,
although I understand the general meaning of contrasts, I do not understand clearly what is the substantial difference with average marginal effects. Both are supposed to measure the difference in the predicted probabilities for different categories and both use a delta-method, isn't it? How to interprete contrasts coefficient wrt dydx coefficients?
Here below an example based on a logit estimating the probability of being self-employed at present as a function of previous statuses: explanatory variables are self in origin, transitions types after origin and the interaction between the two terms.
First I estimate the margins, or the predicted probabilites for each value of the interaction and then I test the difference using contrast.
What additional information the contrast coefficient of (Self-employed vs Employee) unchanged provides wrt the difference between the predicted probabilities of employee#unchanged and self-employed#unchanged? The chi2 test tells us that the difference is significant but I do not know how to interpret the contrast coefficient 1.143266.
margins self_origin#trans_first
Predictive margins Number of obs = 5,807
Model VCE: Robust
Expression: Pr(self_present), predict()
-----------------------------------------------------------------------------------------------
| Delta-method
| Margin std. err. z P>|z| [95% conf. interval]
------------------------------+----------------------------------------------------------------
self_origin#trans_first |
Employee#unchanged | .0828733 .0076741 10.80 0.000 .0678324 .0979142
Employee#first self-emp | .2406017 .0871938 2.76 0.006 .0697049 .4114985
Employee#first employee | .1194083 .0091236 13.09 0.000 .1015263 .1372902
Self-employed#unchanged | .2111513 .0277878 7.60 0.000 .1566882 .2656144
Self-employed#first self-emp | .2570196 .1040233 2.47 0.013 .0531377 .4609015
Self-employed#first employee | .139658 .0225762 6.19 0.000 .0954094 .1839066
-----------------------------------------------------------------------------------------------
contrast r.self_origin@trans_first, effects
Contrasts of marginal linear predictions
Margins: asbalanced
-------------------------------------------------------------------------------
| df chi2 P>chi2
--------------------------------------------+----------------------------------
self_origin@trans_first |
(Self-employed vs Employee) unchanged | 1 31.01 0.0000
(Self-employed vs Employee) first self-emp | 1 0.02 0.9024
(Self-employed vs Employee) first employee | 1 0.75 0.3872
Joint | 3 31.40 0.0000
-------------------------------------------------------------------------------
-------------------------------------------------------------------------------------------------------------
| Contrast Std. err. z P>|z| [95% conf. interval]
--------------------------------------------+----------------------------------------------------------------
self_origin@trans_first |
(Self-employed vs Employee) unchanged | 1.143266 .2053177 5.57 0.000 .7408504 1.545681
(Self-employed vs Employee) first self-emp | .0946731 .7716435 0.12 0.902 -1.41772 1.607067
(Self-employed vs Employee) first employee | .189079 .2186445 0.86 0.387 -.2394563 .6176143
-------------------------------------------------------------------------------------------------------------
Thanks in advance
Floriane
although I understand the general meaning of contrasts, I do not understand clearly what is the substantial difference with average marginal effects. Both are supposed to measure the difference in the predicted probabilities for different categories and both use a delta-method, isn't it? How to interprete contrasts coefficient wrt dydx coefficients?
Here below an example based on a logit estimating the probability of being self-employed at present as a function of previous statuses: explanatory variables are self in origin, transitions types after origin and the interaction between the two terms.
First I estimate the margins, or the predicted probabilites for each value of the interaction and then I test the difference using contrast.
What additional information the contrast coefficient of (Self-employed vs Employee) unchanged provides wrt the difference between the predicted probabilities of employee#unchanged and self-employed#unchanged? The chi2 test tells us that the difference is significant but I do not know how to interpret the contrast coefficient 1.143266.
margins self_origin#trans_first
Predictive margins Number of obs = 5,807
Model VCE: Robust
Expression: Pr(self_present), predict()
-----------------------------------------------------------------------------------------------
| Delta-method
| Margin std. err. z P>|z| [95% conf. interval]
------------------------------+----------------------------------------------------------------
self_origin#trans_first |
Employee#unchanged | .0828733 .0076741 10.80 0.000 .0678324 .0979142
Employee#first self-emp | .2406017 .0871938 2.76 0.006 .0697049 .4114985
Employee#first employee | .1194083 .0091236 13.09 0.000 .1015263 .1372902
Self-employed#unchanged | .2111513 .0277878 7.60 0.000 .1566882 .2656144
Self-employed#first self-emp | .2570196 .1040233 2.47 0.013 .0531377 .4609015
Self-employed#first employee | .139658 .0225762 6.19 0.000 .0954094 .1839066
-----------------------------------------------------------------------------------------------
contrast r.self_origin@trans_first, effects
Contrasts of marginal linear predictions
Margins: asbalanced
-------------------------------------------------------------------------------
| df chi2 P>chi2
--------------------------------------------+----------------------------------
self_origin@trans_first |
(Self-employed vs Employee) unchanged | 1 31.01 0.0000
(Self-employed vs Employee) first self-emp | 1 0.02 0.9024
(Self-employed vs Employee) first employee | 1 0.75 0.3872
Joint | 3 31.40 0.0000
-------------------------------------------------------------------------------
-------------------------------------------------------------------------------------------------------------
| Contrast Std. err. z P>|z| [95% conf. interval]
--------------------------------------------+----------------------------------------------------------------
self_origin@trans_first |
(Self-employed vs Employee) unchanged | 1.143266 .2053177 5.57 0.000 .7408504 1.545681
(Self-employed vs Employee) first self-emp | .0946731 .7716435 0.12 0.902 -1.41772 1.607067
(Self-employed vs Employee) first employee | .189079 .2186445 0.86 0.387 -.2394563 .6176143
-------------------------------------------------------------------------------------------------------------
Thanks in advance
Floriane
