[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
Re: st: XTIVREG models and assumptions
Only a few suggestions:
- unofficial command -abar- from ssc tests for serial correlation
- I have seen on this list some days ago that "another well-kept secret in the area of heteroskedasticity testing is official -robvar-, which is very useful for testing for groupwise heteroskedasticity
- maybe looking at
can provide you some suggestions about how to correct the standard errors.
At 02.33 05/03/2007 -0500, Anand Jeyaraj wrote:
>I am working with a cross-sectional time-series (XT/ panel) dataset of N = 5000 units and T = 50 time peiods. I am interested in examining a model that contains: a) a single dependent variable measured at T time periods, and b) several independent variables, including the lagged dependent variable, a non-linear (squared) variable, a predetermined variable not strictly endogenous, a within-unit time-invariant exogenous variable. I believe that a random-effects model would be appropriate for the model.
>After reviewing the various XT commands availabe in Stata, I believe that XTIVREG would be an appropriate procedure to use for the model as it seems to allow for the various types of independent variables in my model (unlike say, XTGLS or XTABOND that allow only linear relationships). However, I have run into trouble with the tests for specification, stationarity, serial correlation, heteroskedasticity, and normality. From the stata lists and other documentation, I am aware of:
>1. the -hausman- and -xttest0- procedures for specification... which may not be possible after -xtivreg-
>2. the -levinlin- and -xtfisher- procedures for panel stationarity...
>3. the -xtserial- procedure for serial correlation...
>4. the -xttest2- and -xttest3- procedures for heteroskedasticity... which may not be possible for random-effects model
>5. no -xt- procedure for normality. I am aware of the -kdensity- function... which I am not sure is appropriate for examining residuals of panel data. I can't get -xtsum- to display skewness or kurtosis statistics for panel data either.
>I am unsure on the extent to which the five issues above are critical for running an -xtivreg- analysis for my model. I would greatly appreciate your assistance.
>Thank you in advance
* For searches and help try: