Skip to content

linearmodels/hand-written-hausman

Rule hand-written-hausman
Pack linearmodels
Severity info
Fix none, edit by hand
Library versions linearmodels 7.1 and later

linearmodels has both Hausman tests built in, so a hand-written one is extra code to keep right. Call the result's hausman method or the test that the message names.

def hausman(fe, re):
b = fe.params - re.params
v = fe.cov - re.cov
return float(b @ np.linalg.inv(v) @ b)
test = random_effects_result.hausman(fixed_effects_result)

The rule reports a function whose name starts with hausman. It leaves alone hausman_taylor, which is the Hausman and Taylor estimator, and names such as hausman_table or hausman_data, which show results or prepare data. 7.1 added the Hausman test of random effects against fixed effects as a method of the random effects result, and the instrument results have had wu_hausman() for longer.

xpd has no automatic fix. Needs linearmodels 7.1. Make the change by hand.

On 7.1 the method returned a HausmanTestResults with the fields statistic and estimates, and 7.0 has no such method. The rule reports 25 functions in 25 of the 950 public files. It cannot see what a function tests, so it is info, and the message names both tests.

xpd prints this message with the issue. A name that starts with $ stands for code from your file.

linearmodels has both Hausman tests, so $NAME need not be written by hand. For random effects against fixed effects, 7.1 added random_effects_result.hausman(fixed_effects_result), which returns the statistic and the table of estimates. For OLS against instruments, the result of IV2SLS has wu_hausman().

Browse the other rules of the linearmodels pack.