Skip to content

Instantly share code, notes, and snippets.

Embed
What would you like to do?
Summary of the project of panel data estimations of Google Summer of Code 2020.

GSoC 2020 - Panel Data Spatial Econometrics

This post summarizes the work of the PySAL Project on Panel Data Spatial Econometrics. The work is divided in the following sections. First, I explain the utilities used to handle panel data in spreg. Second, I show the diagnostics implemented for spatial - panel estimations. Finally, I detail the different models that can be estimated.

The notebook Panel_example.ipynb offers an overview of the new estimations that can be useful from the user perspective.

Utilities

The utilities for the panel data estimation are located in the file panel_utils.py.

  • The function check_panel handles the structure of the panel data in the estimations of spreg. This function converts a panel from wide to long format if needed.

  • The function demean_panel transforms the variables for the estimations of spreg. The transformation assigns a weight from 0 to 1 attached to the cross-sectional component of the data.

Diagnostics

Diagnostic statistics for the panel data estimation are located in the file diagnostics_panel.py.

  • Lagrange Multiplier test: functions that calculate the classic Lagrange Multiplier test and the robust version for spatial lag and error specifications.

  • Hausman test: functions to test fixed vs. random effects specifications.

Estimation

The four basic estimations of panel data with spatial interactions are located in the files panel_fe.py and panel_re.py.

  • panel_fe.py

    • Panel_FE_Lag: Fixed Effects estimation with spatial lagged dependent variable.

    • Panel_FE_Error: Fixed Effects estimation with spatial error interaction.

  • panel_re.py

    • Panel_RE_Lag: Random Effects estimation with spatial lagged dependent variable.

    • Panel_RE_Error: Random Effects estimation with spatial error interaction.

The step by step explanation of the preceding estimations can be found on the notebooks:

Finally, all the work can be found in the following pull requests:

Future Work

There are three issues outside the scope of the original project that is considered for future work:

  • POLS class for Pooled OLS regresion of panel data.
  • Add the diagnostics tests at the end of POLS summary.
  • Lee-Yu bias correction for fixed effects estimation.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment