Hi,
I've trouble inpterpreting marginal effects in multinomial logistic model with dummy preditors.
I'v already read helpful tech supports in SAS (Usage Note 22604: Marginal effect estimation for predictors in logistic and probit models ) that explained how to calculate marginal effects in logistic model with SAS.
But this manual doesn't treat in case of dummy predictors.
Though I derived marginal effects from 22604 manual, I'm not sure whether it's possible to interpret as continuos predictors case in manual's examples. Is there no problem if I compute and interpret marginal effects in case of dummy predictors and categorical dependent variable as above manul(continous predictors and categorical dependent variable)?
Well, It might be the good way to use odd ratios, but I'd like to use marginal effects if possible.
Thanks for reading.
* Follwing code is about marginal effect of pie1 on c1.
* c1 is dependent variable consist of 0,1,2;
* status is binary dummy predictor : go(good), ba(bad);
* pie is also binay dummy predictor : up, do(down);
data aa;
input c1 $ status $ pie1 $;
cards;
0 ba up
1 ba up
0 ba up
1 ba up
1 ba up
0 ba up
0 ba do
1 ba up
1 ba up
0 ba up
2 go up
1 ba up
0 ba up
1 ba up
1 ba up
1 ba up
1 ba up
2 ba up
0 ba up
1 ba do
;
run;
proc surveylogistic data = aa;
class c1(ref='1') pie1(ref='do') status(ref='ba') / param = ref;
model c1 = pie1 status receive / link = glogit;
ods output ParameterEstimates=betas;
output out=preds predprobs=individual;
run;
proc transpose data=betas out=rowbetas;
var estimate;
run;
data margeff;
if _n_=1 then set rowbetas;
set preds;
SumBetaPred = col3*IP_0 + col4*IP_2 ;
MEc0 = IP_0 * (col3 - SumBetaPred);
MEc2 = IP_2 * (col4 - SumBetaPred);
MEc1 = IP_1 * (-SumBetaPred);
run;
proc sort data = margeff out = margeff2 nodupkey;
by pie1;
run;
data margeff3; set margeff2;
keep pie1 mec1 mec2 mec0 cty;
run;
proc print data = margeff2 ;
id pie1;
var me:;
run;
data margeff3; set margeff2;
keep pie1 mec1 mec2 mec0 cty;
run;
proc print data = margeff2 ;
id pie1;
var me:;
run;
