I'm running Stata 15.1 on OSX and working with yearly panel data. My hypothesis predicts that social media activity (a continuous variable) at t-2 and t-1 will predict agreement with a statement (a binary variable) at t-1 and t. I also aim to test if there's any evidence of a causal relationship, which would, for instance, mean that the effects of social media activity at t-2 on agreement at t-1 are significantly stronger than the effects of agreement at t-2 on social media activity at t-1. In other words, I'd like to test a cross-lagged model. The question relates to the fact that the two variables are measured along different scales--one continuous and the other binary. I'm aware that -sem- only allows for continuous outcome variables, while -gsem- allows for both. If I estimate the model with -gsem-, however, how would I be able to test for equality of coefficients? If I were to instead use -sem-, this issue could be rectified by standardizing the coefficients (i.e. estat stdize: test coefficient1 vs. coefficient2). However, the parameter estimates obtained from using -sem- might be unreliable given the disparately-measured outcome variables. As such, I'm not quite sure how to proceed. Is it possible to standardize and test the equality of coefficients with gsem? If not, should I simply treat the binary variable as continuous and stick to SEM (while leaving open the possibility that the estimates are unreliable?). I would greatly appreciate any advice you might have for approaching this issue.
Here is some sample data:
Code:
* Example generated by -dataex-. To install: ssc install dataex clear input float(sma sma_lag1 sma_lag2) byte whpa float(whpa_lag1 whpa_lag2) double caseid float(time year) -.15783915 . . 0 . . 100197 1 2016 .18903075 -.15783915 . 0 0 . 100197 2 2017 -.13314976 .18903075 -.15783915 1 0 0 100197 3 2018 -.13009462 . . 0 . . 100252 1 2016 0 -.13009462 . 0 0 . 100252 2 2017 0 0 -.13009462 . 0 0 100252 3 2018 -.15783915 . . 0 . . 100260 1 2016 -.15783915 -.15783915 . 0 0 . 100260 2 2017 -.13314976 -.15783915 -.15783915 0 0 0 100260 3 2018 -.06436194 . . 0 . . 100314 1 2016 -.06436194 -.06436194 . 0 0 . 100314 2 2017 -.02402338 -.06436194 -.06436194 0 0 0 100314 3 2018 .02230732 . . 0 . . 100363 1 2016 .56018245 .02230732 . 0 0 . 100363 2 2017 -.13314976 .56018245 .02230732 0 0 0 100363 3 2018 0 . . . . . 100413 1 2016 0 0 . . . . 100413 2 2017 0 0 0 . . . 100413 3 2018 -.15783915 . . 0 . . 100446 1 2016 -.06436194 -.15783915 . 0 0 . 100446 2 2017 -.02402338 -.06436194 -.15783915 0 0 0 100446 3 2018 -.15783915 . . 0 . . 100514 1 2016 -.06436194 -.15783915 . 0 0 . 100514 2 2017 0 -.06436194 -.15783915 . 0 0 100514 3 2018 .02230732 . . 0 . . 100588 1 2016 .02230732 .02230732 . 1 0 . 100588 2 2017 .27179015 .02230732 .02230732 1 1 0 100588 3 2018 -.06436194 . . 0 . . 100598 1 2016 -.06436194 -.06436194 . 0 0 . 100598 2 2017 -.02402338 -.06436194 -.06436194 0 0 0 100598 3 2018 -.13009462 . . 0 . . 100604 1 2016 -.06436194 -.13009462 . 0 0 . 100604 2 2017 -.02402338 -.06436194 -.13009462 0 0 0 100604 3 2018 -.13009462 . . 0 . . 100637 1 2016 -.15783915 -.13009462 . 0 0 . 100637 2 2017 -.13314976 -.15783915 -.13009462 0 0 0 100637 3 2018 .09993055 . . 0 . . 100734 1 2016 .11578453 .09993055 . 0 0 . 100734 2 2017 -.02402338 .11578453 .09993055 0 0 0 100734 3 2018 -.15783915 . . 0 . . 100803 1 2016 -.15783915 -.15783915 . 0 0 . 100803 2 2017 -.13314976 -.15783915 -.15783915 0 0 0 100803 3 2018 0 . . . . . 100866 1 2016 0 0 . . . . 100866 2 2017 0 0 0 . . . 100866 3 2018 0 . . . . . 100982 1 2016 0 0 . . . . 100982 2 2017 0 0 0 . . . 100982 3 2018 -.15783915 . . 0 . . 101224 1 2016 -.15783915 -.15783915 . 0 0 . 101224 2 2017 -.13314976 -.15783915 -.15783915 0 0 0 101224 3 2018 -.15783915 . . 0 . . 101322 1 2016 -.15783915 -.15783915 . 0 0 . 101322 2 2017 -.13314976 -.15783915 -.15783915 0 0 0 101322 3 2018 0 . . . . . 101368 1 2016 0 0 . . . . 101368 2 2017 0 0 0 . . . 101368 3 2018 .25423202 . . 0 . . 101400 1 2016 0 .25423202 . . 0 . 101400 2 2017 .3809165 0 .25423202 . . 0 101400 3 2018 -.15783915 . . 0 . . 101437 1 2016 -.15783915 -.15783915 . 0 0 . 101437 2 2017 -.13314976 -.15783915 -.15783915 0 0 0 101437 3 2018 -.13009462 . . 0 . . 101443 1 2016 0 -.13009462 . . 0 . 101443 2 2017 0 0 -.13009462 . . 0 101443 3 2018 .28250796 . . 0 . . 101472 1 2016 .28250796 .28250796 . 0 0 . 101472 2 2017 .27179015 .28250796 .28250796 0 0 0 101472 3 2018 .4626544 . . 0 . . 101485 1 2016 .4626544 .4626544 . 0 0 . 101485 2 2017 .3809165 .4626544 .4626544 0 0 0 101485 3 2018 -.15783915 . . 0 . . 101493 1 2016 .3967137 -.15783915 . 0 0 . 101493 2 2017 .27179015 .3967137 -.15783915 0 0 0 101493 3 2018 0 . . . . . 101495 1 2016 0 0 . . . . 101495 2 2017 0 0 0 . . . 101495 3 2018 0 . . . . . 101712 1 2016 0 0 . . . . 101712 2 2017 0 0 0 . . . 101712 3 2018 -.15783915 . . 0 . . 102009 1 2016 .11578453 -.15783915 . 1 0 . 102009 2 2017 -.13314976 .11578453 -.15783915 1 1 0 102009 3 2018 0 . . . . . 102130 1 2016 0 0 . 1 . . 102130 2 2017 0 0 0 . 1 . 102130 3 2018 .02230732 . . 0 . . 102198 1 2016 .02230732 .02230732 . 0 0 . 102198 2 2017 -.13314976 .02230732 .02230732 0 0 0 102198 3 2018 -.13009462 . . . . . 102351 1 2016 0 -.13009462 . 0 . . 102351 2 2017 0 0 -.13009462 . 0 . 102351 3 2018 .380036 . . 0 . . 102352 1 2016 0 .380036 . . 0 . 102352 2 2017 0 0 .380036 . . 0 102352 3 2018 0 . . . . . 102400 1 2016 0 0 . . . . 102400 2 2017 0 0 0 . . . 102400 3 2018 -.15783915 . . 0 . . 102584 1 2016 end
-Zach
0 Response to Cross-lagged panel models with binary and continuous outcome variables: gsem or sem (?) + other questions
Post a Comment