Added Variable Plot: Adjusted whole model x-axis range

I created an added variable plot for a whole model, based on 4 predictor variables. I don't understand the values that are now on the x-axis: "adjusted whole model." Where are these numbers actually coming from, and why are they negative?

Thanks~

[jgg]

You need to post examples of the code and what you've done; we can't see your results to interpret them.

[alana2015]

sorry about that! thank you! The axis I am referring to is the x-axis in the bottom graph.

ALL_MONTHS_multilinear_fit=fitlm([ALL_MONTHS_EEMs, ALL_MONTHS_SAL, ALL_MONTHS_pH, ALL_MONTHS_exo_chl], ALL_MONTHS_fDOM)

plot(ALL_MONTHS_multilinear_fit) %where an added variable plot is automatically constructed

Answer 1

[jgg]

The added variable plot is little bit complicated; basically, because you have a multi-variable regression, it is trying to give you a way to visualize the fit of the dependent variable versus the other variables in the model.

However, because there are more than one independent variable, you can't just plot them against one another since that would be misleading. Instead, what it does is use the Frisch–Waugh–Lovell theorem theorem to "partial out" the other variables in the model. In the plot you're seeing, it is partial-ing out all of variables except the constant term and showing you in essence the plot of the constant term once the other variables are controlled for, with the dependent variable also partial-ed out in the same way. There's a good Stackexchange discussion in the first answer here if you want to read more.

You can interpret that graph as showing your model has explanatory power; if it did not, it would appear as a horizontal line in the middle of the graph.

[alana2015]

Thanks for your explanation! I will read that Stackexchange discussion too.

Do you know why the adjusted whole model x-axis would be negative?

[jgg]

Yes; once you control for the other variables, the impact left over by the constant term could be negative. This is an expected result, although it would not occur all the time.

[Emmanuel J Rodriguez]

@jgg thank you for this explanation!

[Regina Katsman]

hi,

could you pls explaine how to get this whole model adjusted Xvalue in the numeric form?

thanks,

Regina

[Regina Katsman]

I meant how to get it from the fitlm function