Hello Statalist,
I am working on a household level model that looks at the impact of climatic conditions on welfare and poverty. My outcome variable is household level total and food expenditure, and my left-hand side variables include household level controls (the standard household size, head's education, assets ownership, ...) and weather data on rainfall and temperature measured at the district level, as following:
EXP_i = X_i + WEATHER_d + error_i
Auffhammer et. al. [1] discuss extensively the main pittfalls of introducing spatial climatic variables into household level regressions. One of them is the existence of spatial correlation among the spatial variables and in the error term that needs to be accounted for. Therefore I have been trying to define a spatial weighting matrix using the user written commands "spmat" and "spwmatrix" and subsequently to run an error correction model using "spreg" and "spmlreg". However, this turns out to be problematic because my GPS coordinates do not uniquely identify my households, since the spatial information is at the district level and I do not have information on the exact location of the households. I think I would need to define a spatial weighting matrix giving the highest weight to all the households in the same district, and progressively lower weights to households in nearby districts.
Are there any suggestions on how to do this? Are there alternative ways to account for the spatial correlation in my left-hand side variables that do not require the construction of spatial weighting matrix / GPS coordinates uniquely identifying the observations? Thank you very much!
[1] Auffhammer, M., Hsiang, S. M., Schlenker, W., & Sobel, A. (2013). Using weather data and climate model output in economic analyses of climate change. Review of Environmental Economics and Policy, ret016.
I am working on a household level model that looks at the impact of climatic conditions on welfare and poverty. My outcome variable is household level total and food expenditure, and my left-hand side variables include household level controls (the standard household size, head's education, assets ownership, ...) and weather data on rainfall and temperature measured at the district level, as following:
EXP_i = X_i + WEATHER_d + error_i
Auffhammer et. al. [1] discuss extensively the main pittfalls of introducing spatial climatic variables into household level regressions. One of them is the existence of spatial correlation among the spatial variables and in the error term that needs to be accounted for. Therefore I have been trying to define a spatial weighting matrix using the user written commands "spmat" and "spwmatrix" and subsequently to run an error correction model using "spreg" and "spmlreg". However, this turns out to be problematic because my GPS coordinates do not uniquely identify my households, since the spatial information is at the district level and I do not have information on the exact location of the households. I think I would need to define a spatial weighting matrix giving the highest weight to all the households in the same district, and progressively lower weights to households in nearby districts.
Are there any suggestions on how to do this? Are there alternative ways to account for the spatial correlation in my left-hand side variables that do not require the construction of spatial weighting matrix / GPS coordinates uniquely identifying the observations? Thank you very much!
[1] Auffhammer, M., Hsiang, S. M., Schlenker, W., & Sobel, A. (2013). Using weather data and climate model output in economic analyses of climate change. Review of Environmental Economics and Policy, ret016.