topic Re: Weighting Observations in Discussions

Weighting Observations

DataMonkeyMan — Thu, 23 May 2019 12:09:35 GMT

I'm trying to pull some summary statistics out of a weighted dataset (e.g., proportion of the population that has blue as a favorite colour). However, JMP does not seem to automatically weight counts. Is there a way to get JMP to produce the weighted count?

Thanks for your help.

Re: Weighting Observations

Mark_Bailey — Thu, 23 May 2019 13:08:49 GMT

Which platform are you using, Distribution? Did you put the data column with the Counts in the Freq analysis role in the launch dialog?

Re: Weighting Observations

DataMonkeyMan — Thu, 23 May 2019 13:13:08 GMT

Yes, I am using the distribution command under Analyze. The variable I'm looking at has been assigned to being nominal, and I've included my weight variable.

Re: Weighting Observations

Mark_Bailey — Thu, 23 May 2019 13:26:38 GMT

Do you have another data column with the weight to be used for each level?

Maybe I should ask what you mean by 'weighted' in this case?

Re: Weighting Observations

DataMonkeyMan — Thu, 23 May 2019 13:34:45 GMT

I have two variables; colour preference and weight. Colour preference are nominal, and weight is continuous. I have a sample of 8,000 that is scaled to a population of 30,000.

Re: Weighting Observations

Mark_Bailey — Thu, 23 May 2019 13:37:16 GMT

Here is an example:

Here is the Distribution analysis:

Alternatively, the levels might be pre-summarized:

But the Distribution analysis is the same using N Rows in the Freq role.

How does weighting enter your analysis?

Re: Weighting Observations

Mark_Bailey — Thu, 23 May 2019 13:39:07 GMT

Ah, then color preference is the Y variable and weight is the Weight variable. There is not count for the Freq role. Does that clarification help?

Re: Weighting Observations

DataMonkeyMan — Thu, 23 May 2019 13:39:28 GMT

I'm wanting to be able to state that X% of the population prefer blue.

Re: Weighting Observations

KarenC — Sun, 26 May 2019 12:23:21 GMT

Try using your weight as the frequency. I am assuming your data tabel has 8000 rows and you have a coulm that "weights" each row to the proprotion of the full population that it represents. So this "wegithing" column is the "frequency" of that response in the population. To get your population counts from your sample use your response (color) as your Y and the "weight" as the FREQ column. Then you will get something like this (my example has N = 28741 rows weighted to a population of N = 955968818)

Raw Data

Level	Count	Prob
1	16751	0.58283
2	11990	0.41717
Total	28741	1.00000

Population Wise:

Level	Count	Prob
1	572697202	0.59908
2	383271616	0.40092
Total	955968818	1.00000