diagnostic plots for regression - another version

[josef-pkt]

see plot at bottom of

http://support.sas.com/documentation/cdl/en/statug/63347/HTML/default/viewer.htm#statug_glm_sect056.htm

for one way analysis

looks better than the one in NIST/dataplot

[josef-pkt]

a nice set of examples using R, mostly car plots
http://rpubs.com/kaz_yos/car-residuals

[josef-pkt]

blog post with nice SAS diagnostic plot after OLS
http://alstatr.blogspot.ca/2015/08/r-python-and-sas-getting-started-with.html

same type of plot as in description

[kdebrab]

@ChadFulton implemented MLEResults.plot_diagnostics and demonstrated its use here.

I think this method can be easily adapted to work for any Results class?

Basically, removing the references to self.loglikelihood_burn, changing line 2052 into resid = self.resid / np.sqrt(np.cov(self.resid)) and line 2060 into ax.plot(x[-len(resid):], resid) seem to do the trick for me.

[josef-pkt]

There are some plots that are specific to time series. The same pattern can be used to attach plot methods to other results, but we have to decide which plots should be included.

E.g. with cross-section data there is no "natural" ordering of observations.

[josef-pkt]

another blog with 4 diagnostic plots to imitate R
https://towardsdatascience.com/going-from-r-to-python-linear-regression-diagnostic-plots-144d1c4aa5a

(I have currently no overview of diagnostic plots, i.e. in graphics, probplots and outlier_influence and maybe some other places and notebooks)

Attaching outlier and influence plot to a results instance would mean that we create the outlier-influence statistics (instance of those classes) but have to throw it away after creating the plot and giving only the plot to users.