def ols_model_diagnostics(model):
#residuals should look normal - this plot should be more of less a straight line
print('Residuals QQ-Plot')
fig = statsmodels.api.qqplot(model.resid,
scipy.stats.t,
fit=True,
line='r')
matplotlib.pyplot.show()
#residuals vs. fitted values - this should look random/no pattern
stdres = pandas.DataFrame(model.resid_pearson)
fig = matplotlib.pyplot.plot(stdres, 'o', ls='none')
l = matplotlib.pyplot.axhline(y=0, color='r')
matplotlib.pyplot.ylabel('Standardized Residual')
matplotlib.pyplot.xlabel('Observation')
print fig
#leverage statistics vs. normalized residuals squared
plot_leverage_resid2(model)
#influence plot: studentized resids vs. leverage. The combination of large residuals and a high leverage (influence on estimation of the model coefficients) indicates an influence point. In both of these plots, mostly checking to see that most points have low leverage
influence_plot(model)
#test for nonlinearity: looking for the p-value here to be > 0.05 to meet the linearity assumption
try:
print linear_harvey_collier(model)
except linalg.LinAlgError:
print "Error: Singular covariance matrix, Harvey Collier nonlinearity test not executed"
def test_harvey_collier(self):
# > hc = harvtest(fm, order.by = NULL, data = list())
# > mkhtest_f(hc, 'harvey_collier', 't')
harvey_collier = dict(statistic=0.494432160939874, pvalue=0.6215491310408242, parameters=(198), distr="t")
# > hc2 = harvtest(fm, order.by=ggdp , data = list())
# > mkhtest_f(hc2, 'harvey_collier_2', 't')
harvey_collier_2 = dict(statistic=1.42104628340473, pvalue=0.1568762892441689, parameters=(198), distr="t")
hc = smsdia.linear_harvey_collier(self.res)
compare_t_est(hc, harvey_collier, decimal=(12, 12))
def test_harvey_collier(self):
#> hc = harvtest(fm, order.by = NULL, data = list())
#> mkhtest_f(hc, 'harvey_collier', 't')
harvey_collier = dict(statistic=0.494432160939874,
pvalue=0.6215491310408242,
parameters=(198), distr='t')
#> hc2 = harvtest(fm, order.by=ggdp , data = list())
#> mkhtest_f(hc2, 'harvey_collier_2', 't')
harvey_collier_2 = dict(statistic=1.42104628340473,
pvalue=0.1568762892441689,
parameters=(198), distr='t')
hc = smsdia.linear_harvey_collier(self.res)
compare_t_est(hc, harvey_collier, decimal=(12, 12))
