-1

After running these command all day, my head is on fire, I am now reaching out. Please don't direct me to papers about Suest that are commonly mentioned on the web, I already checked them. It seems there is a problem with storing the dy/dx values of the AME to merge between different models in the suest command in order to perform the test command.

What I would like to test is if the AME of the lower class/upper class/middle class in one regime/context is statistically significant from the lower class situated in another regime/context. Dependent variable: 3 categories: renter, mortgaged homeownership, outright homeownership.

   *Liberal_market
   mlogit  owner_housing_debt2  United_States United_Kingdom Swizerland  c.age_centered 
   ib0.lower_class ib0.upper_class if homeownership_regimes==1 , 
   baseoutcome(1) 
   margins ,  dydx(lower_class upper_class) coeflegend post
   est store Liberal_market

  *Family_financial_support

 mlogit  owner_housing_debt2  Belgium Finland France Ireland Luxembourg Norway Spain 
 ib0.lower_class ib0.upper_class if homeownership_regimes==2 , baseoutcome(1) 
 margins ,  dydx(lower_class upper_class) coeflegend post
 est store Family_financial_support

 est table Liberal_market  Family_financial_support

 suest Liberal_market Family_financial_support

 **In the end, this is what I want to do:
  test [Liberal_market]1.lower_class =[Family_financial_support]1.lower_class

*error message Liberal_market was estimated with a nonstandard vce (delta) r(322);

-Unfortunately, the following answer from Statalist regarding the nonstandard vce in suest- didn’t help me either https://www.statalist.org/forums/forum/general-stata-discussion/general/1511169-can-not-use-suest-for-margins-after-probit-or-regress

Will appreciate your solution:)

OCR
  • 1
  • 1
  • 1
    I don't think you can use `suest` to combine results created by `margins`. You could combine the results from `mlogit` with `suest`, calculate the margins manually with `xlincom` (from SSC) with the post option, and finally carry out your test. – Wouter Apr 25 '21 at 18:15
  • Dear @Wouter, please see my comment below as an "answer", since I didn't have enough place in the comments. – OCR Apr 25 '21 at 22:37

1 Answers1

0

Thank you. I tried your recommendation. Unfortunately, I could not find an organized document with examples for xlincom with margins. I tried the following code, but my problem is withdrawing the margins of the independent variable categories (1.lower_class and 1.upper_class) from the 2 separate mlogit reg after suest. I mean, how to define in the command that I want 1.lower class from model A and 1.lower_class from model B in the margins and in the xlincom. Please see my example below:

   mlogit  owner_housing_debt2  ib0.lower_class ib0.upper_class if regime==1, 
   baseoutcome(1) 

   est store A

  mlogit  owner_housing_debt2  ib0.lower_class ib0.upper_class if regime==2, 
  baseoutcome(1) 

  est store B

  suest A B


  margins  1.lower_class  1.upper_class, coeflegend post

  lincom _b[1.lower_class] - _b[1.upper_class]
OCR
  • 1
  • 1
  • `suest` stores results as multiple equation models. Syntax to extract an estimate from a certain equation with `xlincom` is `[eqno]exp | [eqno]_b[exp] | eqno:exp`. See [here](https://stackoverflow.com/questions/67151134/null-hypothesis-testing/67153098#67153098) for an example of calculating margins with `lincom` (or `xlincom`). By the way, looking at the example above, an interaction would be easier. I.e.: `mlogit owner_housing_debt2 (ib0.lower_class ib0.upper_class)##regime if inlist(regime,1,2)`. – Wouter Apr 26 '21 at 06:37