I am running a Multinomial, multilevel regression following the "Example 41g — Two-level multinomial logistic regression (multilevel)" guide in Stata.
I four models with different dependent variables.
1. gsem (i.harass_BP4_intervene <- i.gender age i.disability i.victim_di_base i.english i.employment i.education financesat i.partner i.owner i.time_interval_rc i.crime_hot se_basescale M1[individualID]@1 ), mlogit
2. gsem (i.youngpeople_BP4_intervene <-i.gender age i.disability i.victim_di_base i.english i.employment i.education financesat i.partner i.owner i.time_interval_rc i.crime_hot se_basescale M1[individualID]@1 ), mlogit
3. gsem (i.fight_BP4_intervene <- i.gender age i.disability i.victim_di_base i.english i.employment i.education financesat i.partner i.owner i.time_interval_rc i.crime_hot se_basescale M1[individualID]@1 ), mlogit
4. gsem (i.suspicious_BP4_intervene <- i.gender age i.disability i.victim_di_base i.english i.employment i.education financesat i.partner i.owner i.time_interval_rc i.crime_hot se_basescale M1[individualID]@1 ), mlogit
The models work well for the first two variables, but for models 3 and 4 Stata produces the following error. "cannot compute an improvement -- discontinuous region encountered"
If I run these models without M1[individualID]@1 they work. IndividualID is the level 2 indicator. Is this an issue with my model fit? Why does constraining at level 2 cause issues for some models?
I four models with different dependent variables.
1. gsem (i.harass_BP4_intervene <- i.gender age i.disability i.victim_di_base i.english i.employment i.education financesat i.partner i.owner i.time_interval_rc i.crime_hot se_basescale M1[individualID]@1 ), mlogit
2. gsem (i.youngpeople_BP4_intervene <-i.gender age i.disability i.victim_di_base i.english i.employment i.education financesat i.partner i.owner i.time_interval_rc i.crime_hot se_basescale M1[individualID]@1 ), mlogit
3. gsem (i.fight_BP4_intervene <- i.gender age i.disability i.victim_di_base i.english i.employment i.education financesat i.partner i.owner i.time_interval_rc i.crime_hot se_basescale M1[individualID]@1 ), mlogit
4. gsem (i.suspicious_BP4_intervene <- i.gender age i.disability i.victim_di_base i.english i.employment i.education financesat i.partner i.owner i.time_interval_rc i.crime_hot se_basescale M1[individualID]@1 ), mlogit
The models work well for the first two variables, but for models 3 and 4 Stata produces the following error. "cannot compute an improvement -- discontinuous region encountered"
If I run these models without M1[individualID]@1 they work. IndividualID is the level 2 indicator. Is this an issue with my model fit? Why does constraining at level 2 cause issues for some models?
