MATLAB: PlotResiduals looks for multivariant outliers or

linear modelplotresiduals

Hey all!

Does anyone know if plotResiduals (like described here: http://www.mathworks.co.uk/help/stats/linear-regression-model-workflow.html#zmw57dd0e22671) looks for multivariante outliers (univariant: deviation from a normal value on on single variable, multivariante: unusual combination of several variables). I think so, or, as it considers the whole model?

Thank you!

Best Answer

No. In the example they look for outliers in the residuals, not for outliers in the inedpendent variables. That means they look for exceptionally high or low residuals. Detecting an unusual combination of several variables would be much harder and there is no Matlab algorithm for that. If you build a model you are interested in outliers in the residuals, not necessarily in outliers in the dependent variables.

Related Solutions

MATLAB: Changing Bar width of plotResiduals

It is quite strange that mathwork has made it quite difficult to extract data out of plotResiduals function or plot. The following code shows one way. The following code shows one way to change the width of the bars by extracting the data from the plotResiduals figure and then creating a new figure using bar() function.

load carsmall
tbl = table(MPG,Weight);
tbl.Year = categorical(Model_Year);
mdl = fitlm(tbl,'MPG ~ Year + Weight^2');
p = plotResiduals(mdl);
vert = p.Vertices;
vert(1:5:end,:) = [];
hist_data = blockproc(vert, [4 2], @(x) mean(x.data(2:3,:)));
bar(hist_data(:,1), hist_data(:,2), 'BarWidth', 0.7);

MATLAB: Heteroscedasticity: plotResiduals vs archtest

Using the ARCH or GARCH tests depends on what your hypotheses are. I refer you to Box, et al., Time Series Analysis 5th edition, 2016, Section 10.2, ISBN 978-1-118-67502-1. This is something only you can determine, based on your data and what hypotheses you want to test. Also, you’re doing a meta-analysis, with different data sources, so the assumptions of these tests may not apply to your data.

I have the reference, so I scanned that section to see if I could offer some substantive guidance. (I can’t.) I’m not familiar with any of these, because they don’t normally apply to the sort of data I’ve worked with (that tend to be homoscedastic). (I also don’t have the Econometrics Toolbox, since I don’t do econometric analyses, so I have no experience with the functions.)

Best Answer

Related Solutions

MATLAB: Changing Bar width of plotResiduals

MATLAB: Heteroscedasticity: plotResiduals vs archtest

Related Question