Research design

[marnixvandeursen]

Hi,

I try to estimate a model with the use of cross sectional data. My depenent variable is the sales price per hectare of 3370 transactions. Independents are Population growth, Population density, farm density, size and soil quality (categorical: poor, average, good). I divide my sample by estimating the model by year and region, Total number of samples is 4 (2010-2013) plus 2 (northern region and southern region for 2013). I want to test for multicolinearity, heteroskedasticity, spatial dependence (can't find a test for this), normal distribution residuals.

- multicolinearity (cannot find some serious correlations between independents so I assume no multicolinearity).
- heteroskedasticity (this is a problem. Heteroskedasticity robust standard errors cant solve it and still homoskedasticity is rejected on a 0.0000 level).
- Spatial dependence (cannot find a test for this)
- Normal distribution residuals (do not exactly know were to pay attention to when assessing the jarque bera of the residuals. Do also not exactly know if this is necessary with my samples and don't know what my DF are in order to check the t-table,

Maybe I forget some tests which I first need to perform.

I wonder if this is al logic and how I can solve for example heteroskedasticity and check for spatial dependence.

Thanks for you help in advance!

Kr,

Marnix

[startz]

Hi,


- heteroskedasticity (this is a problem. Heteroskedasticity robust standard errors cant solve it and still homoskedasticity is rejected on a 0.0000 level).
-
I wonder if this is al logic and how I can solve for example heteroskedasticity and check for spatial dependence.

If you are using heteroskedastic robust standard errors, why do you think heteroskedasticity is still a problem?