Dear All,
I am testing the curvilinear relationship between cInfComp (level-1 variable) and NPD. I also hypothesize that:
  1. caccfin (level-1 variable) will flatten out the curvilinear relationship
  2. cunav (level-2 variable) will steepen the curvilinear relationship.
Will my model be mis-specified if I do not include cInfComp and cInfComp2 in the model as random slopes? Please, find the Stata code below for further details.
Model 1:
Code:
melogit NPD c.caccfin##c.(cInfComp cInfComp2) c.cunav##c.(cInfComp cInfComp2) lnempl comp i.sector lnage foreign expo i.iso lnexptopman || _all:R.year || Country: cInfComp cInfComp2, difficult
Model 2:
Code:
melogit NPD c.caccfin##c.(cInfComp cInfComp2) c.cunav##c.(cInfComp cInfComp2) lnempl comp i.sector lnage foreign expo i.iso lnexptopman || _all:R.year || Country:, difficult
I should mention that the chi2 for model 1 and 2 are respectively chi2(4) = 5938.20, and chi2(2) = 5769.63. (both significant).

Thank you very much for your support.