What is the best way to test which independent variable has a stronger impact on the dependent variable.

[Dongwuk Kim]

Hi everyone,

One of my hypotheses in my research is about which variable has a bigger impact on the dependent variable.
More specifically I am trying to test whether a firm's internal resources or external conditions would affect more on creative performance and financial performance.
We hypothesized that in the case of creative performance internal resources would matter more and in the case of financial performance, the external conditions would matter more.
Right now my co-author has suggested that we should put both in the model and do a wald test afterward, but I am not sure if that is the best way.
Since we assume both IV interact with each other, I have models with both IV and full model with the interaction of the two.
In both cases of creative performance and financial performance, none of the IV is significant by themselves. However, in the full model with the interaction variable, both are significant as in Wald test was significant.
First of all, am I doing this the right way? is there a better way to do it?

Thank you so much for reading.

Jimmy

[Carlo Lazzaro]

Jimmy:
following the FAQ and posting what you typed and what Stata gave you back worths more than a 1000-word description. Thanks.

[Dongwuk Kim]

Hi Carlo thank you for letting me know. I have done some searching and learned you shouldn't compare IV coefficients when there is an interaction variable.
My next question is that is it okay to use test x y when x and y are very different variables. In my result.

. xtreg change_over_change2 first_product_efficiency atarishock2 logged_rnd_w firmage logged_pubtotal logged_devtotal logged_addon phat_game_logit online_publishing if year<2019 , cluster(id)

.

I want to know if the atarishock2 and first_product_efficiency have a significantly different impact on the dependent variable. But I am not sure if I can use " test atarishock2 =first_product_efficiency " when the IV atarishock2 is a dummy variable and first_product_efficiency is a ratio. I read that two variables have to be the same scale.

[Carlo Lazzaro]

Dongwuk:
the most relevant issue with your outputs is that you do not seem to have a panel-wise effect.
What does -xttest0- tells you?

[Felix Bittmann]

You may want to look into dominance analysis, see
https://psycnet.apa.org/doiLanding?d...2-989X.8.2.129
https://psycnet.apa.org/record/1994-08167-001
https://ideas.repec.org/c/boc/bocode/s457629.html

[Dongwuk Kim]

Hi Carlo,

This is the result of the -xttest0-


Breusch and Pagan Lagrangian multiplier test for random effects

change_over_change2[id,t] = Xb + u[id] + e[id,t]

Estimated results:
| Var sd = sqrt(Var)
---------+-----------------------------
change_~2 | 2.19447 1.481374
e | 1.603956 1.266474
u | .0971737 .311727

Test: Var(u) = 0
chibar2(01) = 671.14
Prob > chibar2 = 0.0000

However, I can't use fixed effects model because one of the IV (first_product_efficiency) is a fixed variable across the panel data.
We are trying to see if the characteristics of the firm'ss first market entry year have an effect across the years.

[Dongwuk Kim]

Thank you so much Felix! I will check dominance analysis!

[Carlo Lazzaro]

Dongwuk:
the -xttest0- actually show evidence of a panel-wise effect.
That said, before discarding -fe- altogether I would consider whether the time-invariant predictor cannot be sacrified.

[Dongwuk Kim]

Thank you for the reply Carlo,

The time-invariant predictor cannot be sacrificed because that is the main IV we are interested in.
We are trying to see if the original conditions of firms can have a constant effect on firms.

Thank you

Jimmy

[Carlo Lazzaro]

Dongwuk:
hence you can stick with -xtreg,re- or take a look to the community-contributed module -mundlak-.