Dear all,
In my (regression) analysis I am trying to see when companies are more likely to demand aid from governments, considering 5 major factors – firms’ size, revenue, unemployment in the economy, imports, gdp growth, and my main explanatory variable “ownership”, that is, if a company is owned by a large multinational corporation.
Some variables vary over time (such as revenue “lrev” and employee size “lnempl”) but others, such as mnc ownership, does not.
My outcome variable is binary – i.e., whether or not a firm has applied for “ad” or not. Main IV is also binary – whether a firm is “mnc_owned” or not. 5 other variables are continuous.
Since the event I am interested in is quite rare, I run a rare events logistic regression using the relogit command (OLS reveal similar results as well).
With all the variables in the model, my results are the following,
relogit ad mnc_owned lrev lnempl unemployment lnimport gdp_growth sector year
Corrected logit estimates Number of obs = 371157
------------------------------------------------------------------------------
| Robust
ad | Coefficient std. err. z P>|z| [95% conf. interval]
-------------+----------------------------------------------------------------
mnc_owned | 2.580213 .083656 30.84 0.000 2.416251 2.744176
lrev | .4812687 .0308463 15.60 0.000 .4208109 .5417264
lnempl | -.1178756 .0324088 -3.64 0.000 -.1813956 -.0543555
unemployment | .0195259 .0905168 0.22 0.829 -.1578838 .1969356
lnimport | -.1665573 .9645765 -0.17 0.863 -2.057093 1.723978
gdp_growth | -.0019636 .0226464 -0.09 0.931 -.0463498 .0424226
sector | .0005393 .0000406 13.29 0.000 .0004598 .0006188
year | .0095729 .0405276 0.24 0.813 -.0698597 .0890055
_cons | -28.80196 95.22616 -0.30 0.762 -215.4418 157.8379
------------------------------------------------------------------------------
The results bear my theoretical expectations.
Now I am trying to provide some visuals with a margins command but I have 2 issues I cannot seem to resolve,
When I try to understand the impact of my main IV (mnc ownership), the response is the following:
margins mnc_owned
factor mnc_owned not found in list of covariates
r(322);
So I decided to use dydx command instead, which seems to work:
margins, dydx (mnc_owned)
Average marginal effects Number of obs = 371,157
Model VCE: Robust
Expression: Linear prediction, predict()
dy/dx wrt: mnc_owned
------------------------------------------------------------------------------
| Delta-method
| dy/dx std. err. z P>|z| [95% conf. interval]
-------------+----------------------------------------------------------------
mnc_owned | 2.580213 .083656 30.84 0.000 2.416251 2.744176
------------------------------------------------------------------------------
But, now I cannot seems to see the impact of other continuous variables, such as revenue or size at different levels.
For revenue, for instance, when I try,
margins, dydx(lrev) at(lrev=1)
Average marginal effects Number of obs = 371,157
Model VCE: Robust
Expression: Linear prediction, predict()
dy/dx wrt: lrev
At: lrev = 1
------------------------------------------------------------------------------
| Delta-method
| dy/dx std. err. z P>|z| [95% conf. interval]
-------------+----------------------------------------------------------------
lrev | .4812687 .0308463 15.60 0.000 .4208109 .5417264
------------------------------------------------------------------------------
But for margins, dydx(lrev) at(lrev=6) the result is also identical – which does not make any sense… the level of revenue between 1 and 6 should most certainly not reveal identical results...
1) I cannot seem to see the difference in the different levels of this variable – how can I see the marginal effect of revenue at 1 versus revenue at 7?
2) Is it also possible to do this by keeping another variable in a given value, such as
margins, dydx(mnc_owned) at(lrev=1)
When I try this, the results do not change whether revenue (lrev) is 1 or 3 or 5 or 6 ... I am trying to understand what I am doing wrong.
I hope I was able to formulate these two issues.
Any help is appreciated.
Best,
Aydin
In my (regression) analysis I am trying to see when companies are more likely to demand aid from governments, considering 5 major factors – firms’ size, revenue, unemployment in the economy, imports, gdp growth, and my main explanatory variable “ownership”, that is, if a company is owned by a large multinational corporation.
Some variables vary over time (such as revenue “lrev” and employee size “lnempl”) but others, such as mnc ownership, does not.
My outcome variable is binary – i.e., whether or not a firm has applied for “ad” or not. Main IV is also binary – whether a firm is “mnc_owned” or not. 5 other variables are continuous.
Since the event I am interested in is quite rare, I run a rare events logistic regression using the relogit command (OLS reveal similar results as well).
With all the variables in the model, my results are the following,
relogit ad mnc_owned lrev lnempl unemployment lnimport gdp_growth sector year
Corrected logit estimates Number of obs = 371157
------------------------------------------------------------------------------
| Robust
ad | Coefficient std. err. z P>|z| [95% conf. interval]
-------------+----------------------------------------------------------------
mnc_owned | 2.580213 .083656 30.84 0.000 2.416251 2.744176
lrev | .4812687 .0308463 15.60 0.000 .4208109 .5417264
lnempl | -.1178756 .0324088 -3.64 0.000 -.1813956 -.0543555
unemployment | .0195259 .0905168 0.22 0.829 -.1578838 .1969356
lnimport | -.1665573 .9645765 -0.17 0.863 -2.057093 1.723978
gdp_growth | -.0019636 .0226464 -0.09 0.931 -.0463498 .0424226
sector | .0005393 .0000406 13.29 0.000 .0004598 .0006188
year | .0095729 .0405276 0.24 0.813 -.0698597 .0890055
_cons | -28.80196 95.22616 -0.30 0.762 -215.4418 157.8379
------------------------------------------------------------------------------
The results bear my theoretical expectations.
Now I am trying to provide some visuals with a margins command but I have 2 issues I cannot seem to resolve,
When I try to understand the impact of my main IV (mnc ownership), the response is the following:
margins mnc_owned
factor mnc_owned not found in list of covariates
r(322);
So I decided to use dydx command instead, which seems to work:
margins, dydx (mnc_owned)
Average marginal effects Number of obs = 371,157
Model VCE: Robust
Expression: Linear prediction, predict()
dy/dx wrt: mnc_owned
------------------------------------------------------------------------------
| Delta-method
| dy/dx std. err. z P>|z| [95% conf. interval]
-------------+----------------------------------------------------------------
mnc_owned | 2.580213 .083656 30.84 0.000 2.416251 2.744176
------------------------------------------------------------------------------
But, now I cannot seems to see the impact of other continuous variables, such as revenue or size at different levels.
For revenue, for instance, when I try,
margins, dydx(lrev) at(lrev=1)
Average marginal effects Number of obs = 371,157
Model VCE: Robust
Expression: Linear prediction, predict()
dy/dx wrt: lrev
At: lrev = 1
------------------------------------------------------------------------------
| Delta-method
| dy/dx std. err. z P>|z| [95% conf. interval]
-------------+----------------------------------------------------------------
lrev | .4812687 .0308463 15.60 0.000 .4208109 .5417264
------------------------------------------------------------------------------
But for margins, dydx(lrev) at(lrev=6) the result is also identical – which does not make any sense… the level of revenue between 1 and 6 should most certainly not reveal identical results...
1) I cannot seem to see the difference in the different levels of this variable – how can I see the marginal effect of revenue at 1 versus revenue at 7?
2) Is it also possible to do this by keeping another variable in a given value, such as
margins, dydx(mnc_owned) at(lrev=1)
When I try this, the results do not change whether revenue (lrev) is 1 or 3 or 5 or 6 ... I am trying to understand what I am doing wrong.
I hope I was able to formulate these two issues.
Any help is appreciated.
Best,
Aydin
