Where can I find official documentation on the working mechanisms of Weight and Freq in Fit Y by x

[#1]

Hello engineers!

There are no specific instructions on how to use Weight and Freq in Fit Y by X in JMP's documentation and community, and the explanations online are also very messy.

Is there an official document that provides more specific instructions on how these two parameters work during the fitting process in JMP？ Thank you all for your help all along!

[#2]

here is a collection of findings: Not sure how to use Weight in Analyze 

Here is a wish to improve the documentation for JMP:

Freq and Weight: better documentation 

e.g. via a white paper - like it's available for SAS.

Important:
Highlight the difference to SAS.

[#3]

Hi @BigBook,

You can use Weight and Freq roles in many JMP platforms.

A description of the role and impact of these two roles can be found in JMP Help : 

Elements in the Fit Model Launch Window

Launch the Bivariate Platform 

And a SAS blog on the distinction between the two roles : https://blogs.sas.com/content/iml/2013/09/13/frequencies-vs-weights-in-regression.html 

Observations with higher weight contribute more heavily to the loss function/model fitting than rows involving a smaller weight. Weight has an impact on the loss function and model fitting (change in parameters estimates), and slight changes are possible for statistical significance of the model and effect terms depending on the change in model adequacy between the "normal" model and the "weighted" model. 

"Assigning a frequency is useful when your data are summarized" : The Freq role will affect the number of rows/observations used in the modeling, so it has an impact on the loss function and model fit/results (parameters estimates), but also on degrees of freedom available for pure error estimation, and p-values for statistical model significance and effects significance.

I just came across this situation yesterday as I was trying to make sense of data evaluated differently, without changing the number of observations used. You can find the comparative study I have done on my dataset, where I fit the same model using no Weight/Freq role, a Weight role, and a Freq role. Significant changes are circled in red :

In my case, weight was a better option, as it keeps the number of observations unchanged, but put more emphasis on observations with greater confidence/accuracy.

So you can see that weight and freq have conceptually different roles, and practically a different impact on the model.

 Hope this answer may help you,

[#4]

Thank you for answering my questions in combination with your own exploration, I will study first, and then ask you if you have questions。

thank you very much!

[#5]

I believe these various references are all correct, but they don't quite address an important question.  I often try to analyze survey data from Federal agencies.  These always provide a set of weights - they usually use complex stratified sampling so the weights are important.  Frequency usually suffices for getting descriptive or summary data.  But inference is a different matter and I believe JMP does not provide the correct inference for such cases (I think SAS does - at least that is the answer I've gotten in the past).  I don't know of a way for JMP to provide the correct standard errors in such cases using either the frequency or weight role.

If what I said is correct, it would be helpful to clarify that this is indeed the case.  It would be even better if JMP enabled producing the correct standard errors from these sampling design weights.