linearmodels/absorb-not-categorical
| Rule | absorb-not-categorical |
|---|---|
| Pack | linearmodels |
| Severity | info |
| Fix | none, edit by hand |
| Library versions | every version |
AbsorbingLS absorbs a column as fixed effects only when its dtype is categorical, so a numeric id becomes one continuous regressor. Cast the column to category before you pass it.
model = AbsorbingLS(data.y, data[["x"]], absorb=data[["firm"]])model = AbsorbingLS(data.y, data[["x"]], absorb=data[["firm"]].astype("category"))What the rule reports
Section titled “What the rule reports”The rule reports AbsorbingLS(..., absorb=...) in a file that never says category,
Categorical or cut(. absorb reads the dtype of each column. A categorical column is absorbed
as one effect for each category, and any other column as one continuous regressor.
How to fix it
Section titled “How to fix it”xpd has no automatic fix, so make the change by hand.
With 200 firms, a firm effect in the regressor and a true coefficient of 1.0, no form raised an error or a warning:
absorb |
Coefficient |
|---|---|
| the integer firm id | 3.397 |
| none | 3.398 |
| the firm id as a category | 1.022 |
A text column raises TypeError on 7.1, and 7.0 took it like the integers. One of the 4 public files
that the rule reports turns its columns into text with astype(str) right before the call. The
code doesn't show a dtype, so the rule is info. Data that is categorical when it is loaded is
reported without need.
Message
Section titled “Message”xpd prints this message with the issue. A name that starts with $ stands for code from your file.
AbsorbingLS absorbs a column as fixed effects only when its dtype is categorical. A numeric column is absorbed as one continuous regressor. With an integer firm id the coefficient was 3.397, the same as with no absorb, against 1.022 with the id as a category, where the true value is 1.0. This file never makes a categorical, so check the dtypes, or pass $X.astype("category").Browse the other rules of the linearmodels pack.