topic Re: Goodness of fit test in SAS Studio

Goodness of fit test

laurenhosking — Mon, 30 Dec 2019 21:34:18 GMT

I have a set of data imported how would I perform I goodness of fit test on certain columns of the table. I don’t know where to start.

Re: Goodness of fit test

PaigeMiller — Mon, 30 Dec 2019 21:46:35 GMT

"Goodness-of-fit" test needs context. Without that context, there is no answer.

Sometimes you are testing to see how well data comes from a theoretical distribution (such as the normal distribution). Other times you are testing how well a statistical model fits the data.

Re: Goodness of fit test

ballardw — Mon, 30 Dec 2019 21:46:49 GMT

What kind of goodness of fit?

Against a known distribution such as "Is my variable normally distributed with specific mean and standard deviation?" or similar information for other distributions.

Or "Is my variable A in the data distributed similar to Variable X in this data?"

Re: Goodness of fit test

laurenhosking — Mon, 30 Dec 2019 21:50:32 GMT

I’m honestly not sure I think that’s the issue I’m having but the question says
” Use Goodness of fit test to establish if height (ht) and weight (wt) were
sample from a normal distributed population”

Re: Goodness of fit test

laurenhosking — Mon, 30 Dec 2019 21:50:57 GMT

’m honestly not sure I think that’s the issue I’m having but the question says
” Use Goodness of fit test to establish if height (ht) and weight (wt) were
sample from a normal distributed population”

Re: Goodness of fit test

PaigeMiller — Mon, 30 Dec 2019 22:02:50 GMT

The NORMAL option of PROC UNIVARIATE will cause a goodness-of-fit to the normal distribution to be computed.

https://documentation.sas.com/?docsetId=procstat&docsetTarget=procstat_univariate_syntax01.htm&docsetVersion=9.4&locale=en#procstat.univariate.proc_normal

Re: Goodness of fit test

laurenhosking — Mon, 30 Dec 2019 22:08:19 GMT

Brilliant. Is there an option in sas studio to do this without typing in any form of coding?

Re: Goodness of fit test

PaigeMiller — Mon, 30 Dec 2019 22:16:31 GMT

I can't answer SAS Studio questions if they do not involve writing code. I simply don't use SAS Studio for any other purpose than writing code.

Re: Goodness of fit test

laurenhosking — Mon, 30 Dec 2019 22:29:41 GMT

If you can explain it in code that would help also. I’m just used to using the task bar for most things however I’m open to learning the code

Re: Goodness of fit test

ballardw — Tue, 31 Dec 2019 15:57:16 GMT

@laurenhosking wrote:
If you can explain it in code that would help also. I’m just used to using the task bar for most things however I’m open to learning the code

Basic code to test for normal distribution of one or more variables.

proc univariate data=sashelp.class normal;
   var height weight;
run;

The NORMAL option on the proc statement requests the tests for normality of the variables on the Var statement.

You will get the typical Univariate output of summary moments, statistics, quantiles and extreme plus a table with tests for normality.

Re: Goodness of fit test

laurenhosking — Tue, 31 Dec 2019 16:16:06 GMT

Hi this is really helpful! Just to clarify, I’ll have one set of table for height one set of tables for weight

Re: Goodness of fit test

ballardw — Tue, 31 Dec 2019 18:44:13 GMT

@laurenhosking wrote:

Hi this is really helpful! Just to clarify, I’ll have one set of table for height one set of tables for weight

Yes, each variable on the VAR statement get's separate output in the results. It will be preceded with something like:

Variable: Height

Or you can direct output to data sets but that may be more confusing at the moment.

Re: Goodness of fit test

laurenhosking — Tue, 31 Dec 2019 18:46:16 GMT

Thank you I did this. How does one then interoperate the output

Re: Goodness of fit test

laurenhosking — Wed, 01 Jan 2020 13:31:14 GMT

@PaigeMiller I manages to do this test and get an output with 3 different tests and there p vales. How would I interoperate this to establish wether it not they were a sample from a normal distribution.

Re: Goodness of fit test

PaigeMiller — Wed, 01 Jan 2020 13:48:13 GMT

Show us the output you get.

Re: Goodness of fit test

laurenhosking — Wed, 01 Jan 2020 13:52:52 GMT

Re: Goodness of fit test

PaigeMiller — Wed, 01 Jan 2020 13:57:54 GMT

SAS provides three different tests for goodness-of-fit to a normal distribution, the Kolmogorov-Smirnov, Cramer-von-Mises and the Anderson-Darling. Each tests to see if the hypothesis that the data is from a normal distribution, or not. While I can't really explain the differences, they all say the same thing ... since the p-values are >0.05, you do not reject the hypothesis that the data came from a normal distribution (or in layman's terms, the data is consistent with a normal distribution). You can read more about Kolmogorov-Smirnov here: https://blogs.sas.com/content/iml/2019/05/15/kolmogorov-d-statistic.html

Re: Goodness of fit test

laurenhosking — Sun, 05 Jan 2020 12:07:51 GMT

Brilliant thank you. If I have 2 less than 0.05 but one bigger would I just assume the 2 which are less than?

@PaigeMiller