Dear learned scholars,
I am trying to estimate the effect of parents homeownership on academic performance of their children between 8-13 years old. Meaning does parents having legal property rights over their house influence how their children perform at schools. I have survey data from 4 waves carried out at interval of 2 years. So children drop out of this survey when beyond age 13 and new ones come in when they reach 8 years. I have read here on how to do panel data analysis but the ideas seems to be mixed. I have a few clarifications to make.
1. If I want to make use of data from all the waves (I will unbalanced panel), will using ‘xtreg’ and ‘fe vce(robust)’ be adequate.
2. What generalisations I will not be able to make as I have unbalanced panel as opposed to having balanced panel.
3. Should I also use time dummy and regional dummy in my regression?
4. What diagnostic tests should I do after getting my estimation results?
5. Will running separate cross-section analysis be a better option as opposed to panel data analysis.
Thanks in advance.
Unbalanced panel: when some observations
cross-section or thunderstorms are missing
How to do if there are missing values?
i) Average on two or more observations close to the missing value, ii) if I have only an empty space I can put the same number before or after ... .iii) OLS interpolation iv) I leave blank I do not consider the missing observation if the program allows it ... ...
Thanks Emilio, would you have any reference to any material which shows how it is done like, any youtube video or paper
1) Apart from using a fixed effects estimator, you should also consider using a random effects estimator. Then you can use the Hausman Test to help you decide which estimator is more suitable.
2) If you have enough observations in the data to run your regressions, you do not need to worry much about generalizations.
3) Yes, it is important you include the time and regional dummies because they help you control for region-specific heterogeneity(fixed effects) as well as time-varying effects.
4) If you run both the RE and FE models, the Hausman test is inevitable. However, you can also use system GMM estimator. This will help you deal with the problem of endogeneity, which arises because of reverse causality, omitted variables or measurement errors. For system GMM, you will have conduct the following tests: Arellano Bond test for serial correlation and the Hansen test for overidentifying restrictions.
5) Running cross sectional regressions can never be better than panel regressions. This is because, according to Baltagi (2013) panel data has the following merits;
a) It enables us control for individual heterogeneity
b)Panels give more informative data, more variability, less collinearity among the variables, more degrees of freedom and more efficiency
c) Panels allow for the analysis of dynamics
- Badges
- Science topic
- Similar topics
- Medicine
- Epi
- Age Groups
- Children