Unexpected results from LinearModelFit

Questions
, ,
1

I’m struggling to understand why Mathematica 13.1 gives multiple linear regression coefficients that are greatly different from the Data Analysis package in Excel and Maple’s LinearFit function.

I’m assisting a colleague who is using Excel to find a model for data with 4 independent variables and 1 response variable. All she had access to was Excel, and I attempted to extol the virtues of Mathematica.

Assuming she had selected a reasonable approach, (exponential fitting by transforming the response variable to ln(y)), I tried to duplicate that effort in Mathematica. (Then I would go on to try nonlinear regression.)

But I came up with regression coefficients that were vastly different. I assumed Excel was wrong, so I used Maple to confirm. Nope. Maple agrees with Excel. Even the multiple linear regression calculator on Statistics Kingdom agrees with Excel and Maple. https://www.statskingdom.com/410multi_linear_regression.html

Notebook is attached. Please see the “subsubsection” entitled “exponential”. A screenshot from Excel is embedded for comparison. I used the “ParameterTable” option to show the coefficients.

So, I think that Mathematica is reporting what I don’t understand. Does anyone have any ideas?

Thanks!

raw data.xlsx (39.4 KB)
regression modeling.nb (607 KB)

2

Hi Glen,

There is an error during the evaluation of dataTransformed so it is not transformed.

dataTransformed == data
(* True *)

Try

dataTransformed = ReplaceAt[data, y_ :> Log@y, {All, 5}] /. Indeterminate -> 0
3

Thank you!

I didn’t even think of using ReplaceAt. I had thought Delete then AppendTo was changing the list.

Well, heck!

That was simple…