topic Re: How to simply interpret the result of GLM? in Discussions

How to simply interpret the result of GLM?

JK — Sat, 10 Jun 2023 23:34:52 GMT

Hello, this is a simple question about how to interpret the result of GLM. This is my data.

and I used GLM analysis.

What I understood in GLM, the estimated GN in CV1 under CTRL is

y = 22876.914 - 1901.937 + 900.39475 - 346.4365

= 21528.93

then, the estimated GN in CV2 under CTRL is

y = 22876.914 - 0 + 900.39475 - 0

= 23777.31

---

the estimated GN in CV1 under CTRL is correct compared with actual mean of CV1/CTRL

(23215+23653+17719)/3 = 21529

but, the estimated GN in CV1 under CTRL is not the same as actual mean of CV2/CTRL

(24515+28260+25302)/3 = 26025

What I did misunderstand about GLM? Could you check about this?

Thanks

Re: How to simply interpret the result of GLM?

Mark_Bailey — Fri, 06 Aug 2021 16:52:42 GMT

The report hides the estimates for half of the parameters. The reason is because the estimates for all the parameter levels must sum to zero in this parameterization of the model. Click the red triangle at the top and select Estimates > Expanded Estimates to see the missing values.

Re: How to simply interpret the result of GLM?

JK — Fri, 06 Aug 2021 19:05:17 GMT

Dear. Mark Bailey

Thank you for your answer. Could you indicate me which one do I need to select?

I can't find 'Estimates' in the output. I attach the screenshot.

Could you check about this? I use JMP Pro 15 version.

Many thanks!!

Sincerely,

Re: How to simply interpret the result of GLM?

Mark_Bailey — Mon, 16 Aug 2021 14:49:46 GMT

I am sorry. I was thinking of the ordinary least squares regression provided by Fit Least Squares platform. The GLM platform does not provide this option. The JMP parameterization is such that the estimate for the missing parameter is equal to the negative of the sum of the other estimates.

You might find the Prediction Profiler a convenient way to understand the model predictions. You can see the command to open it in the picture you included. You can click or drag to change the factor levels to obtain a prediction (point and interval estimates).