Dear readers,
For some time I have been busy with a project of mine. In this project I am encountering multiple difficulties on which I have been stuck for some time. In short, the project looks into how youth care demand in children depends on teacher shortages, and in turn how these teacher shortage depend on school characteristics. I arrived at these equations:
EQ1: YouthCareDemand = TeacherShortage + HouseholdVariables + error1
EQ2: TeacherShortage = SchoolCharactersitics + error2
At first I was looking into 2SLS, but 3SLS seems more better because the simultaneous estimation of 3SLS would at least allow the error terms in the two equations to be correlated. This feels more appropriate because I fear (1) omitted variables in the second equation influence both youth care demand and teacher shortages and (2) youth care demand in children might influence teacher shortages. The dataset contains panel data on individuals (some 20.000, over 3 years) and their youth care use, household variables and school characteristics (+TeacherShortage). This model specification raises a few questions for me (of which I would already appreciate only one being answered!):
Q1: Does this seem like a appropriate specification?
Q2: I looked into the manual of the reg3 command on Stata, but I am confused on how to exactly specify such a model. In particular, could I incorperate fixed- and/or random effects into (one of) the two equations?
Q3: Additionally, there is the problem that some school are more represented by individuals in the dataset then others (that is, EQ2 is theoretically an equation at the school-level, while EQ1 at the individual level). How do I account for this? Can I make one equation at the individual level, and the other at the school level? Or does reg3 support weighting in my case? Or should we manually do it (I already have a frequency-weight variable).
Q4: If I wanted to make the TeacherShortage variable a latent variable, captured by 2 or 3 school characteristics, does reg3 support this? or do we need to use other methods (such as the sem() command)
Q5 Lastly does it make sense to specify a few of the SchoolCharacteristics as endogenous in such a model?
Many thanks in advance!
For some time I have been busy with a project of mine. In this project I am encountering multiple difficulties on which I have been stuck for some time. In short, the project looks into how youth care demand in children depends on teacher shortages, and in turn how these teacher shortage depend on school characteristics. I arrived at these equations:
EQ1: YouthCareDemand = TeacherShortage + HouseholdVariables + error1
EQ2: TeacherShortage = SchoolCharactersitics + error2
At first I was looking into 2SLS, but 3SLS seems more better because the simultaneous estimation of 3SLS would at least allow the error terms in the two equations to be correlated. This feels more appropriate because I fear (1) omitted variables in the second equation influence both youth care demand and teacher shortages and (2) youth care demand in children might influence teacher shortages. The dataset contains panel data on individuals (some 20.000, over 3 years) and their youth care use, household variables and school characteristics (+TeacherShortage). This model specification raises a few questions for me (of which I would already appreciate only one being answered!):
Q1: Does this seem like a appropriate specification?
Q2: I looked into the manual of the reg3 command on Stata, but I am confused on how to exactly specify such a model. In particular, could I incorperate fixed- and/or random effects into (one of) the two equations?
Q3: Additionally, there is the problem that some school are more represented by individuals in the dataset then others (that is, EQ2 is theoretically an equation at the school-level, while EQ1 at the individual level). How do I account for this? Can I make one equation at the individual level, and the other at the school level? Or does reg3 support weighting in my case? Or should we manually do it (I already have a frequency-weight variable).
Q4: If I wanted to make the TeacherShortage variable a latent variable, captured by 2 or 3 school characteristics, does reg3 support this? or do we need to use other methods (such as the sem() command)
Q5 Lastly does it make sense to specify a few of the SchoolCharacteristics as endogenous in such a model?
Many thanks in advance!
