Possibilities of adding dummy variables to spatial panel model?

[shuai-zhou]

If my understanding is right, the spatial model specification needs the dataset to be in "wide" format, for example, the fixed effects spatial lag model, then how can I add a dummy variable, for instance, a year dummy into the model? Thanks.

[pedrovma]

Hi @shuai-zhou , I've just updated the FE_Panel notebook to show how you can use data in both formats, 'wide' or 'long'. The function accepts both types of data. I hope this helps.

[shuai-zhou]

Hi, @pedrovma,

Thank you so much for your prompt response. The "baltimore" data is in itself a cross-sectional dataset, I think the "NAT" data can be a good example in implementing the panel model with year dummy variables. And the "long" data format looks like in the following table. My question is, how can I fit a fixed effects spatial lag model with year dummy variable like: hr ~ rd + ps + year_dum. I compiled the "long" data format for you HERE, feel free to compile the data that you think will do the work. Thanks.

name	fips	fipsno	hr	rd	ps	geometry	year	year_dum
Lake of the Woods	27077	27077	0.000000	-0.196536	-1.462559	POLYGON ((...))	1970	1
Ferry	53019	53019	0.000000	-0.847856	-1.697720	POLYGON ((...))	1970	1
Stevens	53065	53065	1.915158	-0.225283	-0.591883	POLYGON ((...))	1970	1
Okanogan	53047	53047	1.288643	-0.391126	-0.552016	POLYGON ((...))	1970	1
Pend Oreille	53051	53051	0.000000	-0.451457	-1.181754	POLYGON ((...))	1970	1
...	...	...	...	...	...	...	...	...
Lake of the Woods	...	...	...	...	...	...	1980	2
Ferry	...	...	...	...	...	...	1980	2
Stevens	...	...	...	...	...	...	1980	2
Okanogan	...	...	...	...	...	...	1980	2
Pend Oreille	...	...	...	...	...	...	1980	2
...	...	...	...	...	...	...	...	...
Lake of the Woods	...	...	...	...	...	...	1990	3
Ferry	...	...	...	...	...	...	1990	3
Stevens	...	...	...	...	...	...	1990	3
Okanogan	...	...	...	...	...	...	1990	3
Pend Oreille	...	...	...	...	...	...	1990	3

[pedrovma]

Hi @shuai-zhou ,

You can just add the dummies as individual X variables for each of the years in your data (minus 1, the reference category).

Example:

import libpysal
import spreg
import geopandas as gpd
import pandas as pd

data = gpd.read_file('nat_long.shp')
data = pd.get_dummies(data, columns=['year']) #This will create the dummies in the dataframe

y = data[['hr']]
x = data[['rd','ps','year_1980','year_1990']] #year_1970 will be the reference category here

w = libpysal.weights.KNN.from_dataframe(data.iloc[0:3085,:],k=10) #W must still be a NxN matrix.
w.transform = 'r'

fe_lag = spreg.Panel_FE_Lag(y.to_numpy().reshape((data.shape[0],1)), x.to_numpy(),
                            w, name_y=list(y.columns), name_x=list(x.columns), name_ds="nat_long.shp")
print(fe_lag.summary)

[shuai-zhou]

Hi, @pedrovma:

Awesome! This is a great example of implementing spatial panel models with dummy variables. Thanks.