Hello all,
I have panel data and use -xtreg, fe- for a regression with Tobin's q as dependent variable. The independent variable is a binary variable (purpose).
I now want to include a moderator which is also a binary variable (b2c). My code is this (with 3-year time lag):
xtreg tq purpose_l3##b2c_l3 cf_l3 growth_l3 pcount_l3 capex_l3 adex_l3 os_l3 fs_l3 i.fyear, fe vce(cluster gvkey)
The regression runs through with the single effect of b2c omitted.
If I want to plot the interaction effect, I use this command:
margins purpose_l3##b2c_l3
marginsplot
Unfortunately, I then only get values for purpose = 1 and = 0 out. I get no values for b2c = 1 and = 0 and for the interaction effect:
-----------------------------------------------------------------------------------
| Delta-method
| Margin std. err. z P>|z| [95% conf. interval]
------------------+----------------------------------------------------------------
purpose_l3 |
0 | 1.838104 .0062248 295.29 0.000 1.825903 1.850304
1 | 2.389538 .2306811 10.36 0.000 1.937411 2.841665
|
b2c_l3 |
0 | . (not estimable)
1 | . (not estimable)
|
purpose_l3#b2c_l3 |
0 0 | . (not estimable)
0 1 | . (not estimable)
1 0 | . (not estimable)
1 1 | . (not estimable)
-----------------------------------------------------------------------------------
Accordingly, the graph is also not useful.
My question is whether there is a solution for this. Should I perhaps use the regression with -xtreg, re- instead of -xtreg, fe-? Would I also have to check industry fixed effects in that case (i.sic)?
Is the problem that the two interacting variables are binary?
I would be very grateful for any assistance!
Kind regards,
Jana
I have panel data and use -xtreg, fe- for a regression with Tobin's q as dependent variable. The independent variable is a binary variable (purpose).
I now want to include a moderator which is also a binary variable (b2c). My code is this (with 3-year time lag):
xtreg tq purpose_l3##b2c_l3 cf_l3 growth_l3 pcount_l3 capex_l3 adex_l3 os_l3 fs_l3 i.fyear, fe vce(cluster gvkey)
The regression runs through with the single effect of b2c omitted.
If I want to plot the interaction effect, I use this command:
margins purpose_l3##b2c_l3
marginsplot
Unfortunately, I then only get values for purpose = 1 and = 0 out. I get no values for b2c = 1 and = 0 and for the interaction effect:
-----------------------------------------------------------------------------------
| Delta-method
| Margin std. err. z P>|z| [95% conf. interval]
------------------+----------------------------------------------------------------
purpose_l3 |
0 | 1.838104 .0062248 295.29 0.000 1.825903 1.850304
1 | 2.389538 .2306811 10.36 0.000 1.937411 2.841665
|
b2c_l3 |
0 | . (not estimable)
1 | . (not estimable)
|
purpose_l3#b2c_l3 |
0 0 | . (not estimable)
0 1 | . (not estimable)
1 0 | . (not estimable)
1 1 | . (not estimable)
-----------------------------------------------------------------------------------
Accordingly, the graph is also not useful.
My question is whether there is a solution for this. Should I perhaps use the regression with -xtreg, re- instead of -xtreg, fe-? Would I also have to check industry fixed effects in that case (i.sic)?
Is the problem that the two interacting variables are binary?
I would be very grateful for any assistance!
Kind regards,
Jana
