LISTSERV at the University of Georgia
Menubar Imagemap
Home Browse Manage Request Manuals Register
Previous messageNext messagePrevious in topicNext in topicPrevious by same authorNext by same authorPrevious page (April 2006, week 3)Back to main SAS-L pageJoin or leave SAS-L (or change settings)ReplyPost a new messageSearchProportional fontNon-proportional font
Date:         Thu, 20 Apr 2006 05:53:56 -0700
Reply-To:     Yung-jui Yang <yungjui@GMAIL.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Yung-jui Yang <yungjui@GMAIL.COM>
Organization: http://groups.google.com
Subject:      Re: SAS macro for testing moderation/interaction
Comments: To: sas-l@uga.edu
In-Reply-To:  <BAY103-F32A66653F9A71C9754E14CB0BA0@phx.gbl>
Content-Type: text/plain; charset="iso-8859-1"

Hi David,

> If you have two continuous regressors, then ODS Statistical Graphics does something else for you. But if you have two continuous regressors, then you ought to be using PROC REG (or something like it) instead of PROC GLM.

True, but ODS Graphics isn't cure-all. As shown in http://support.sas.com/onlinedoc/913/getDoc/zt/statug.hlp/reg_sect50.htm ODS Graphics in PROC REG doesn't plot interactions, whereas my macros can plot them easily.

> Procs like PROC IML take a *massive* hit because they're trying to store everything in RAM instead of using the I/O sub-systems that are the underpinnings of the stat and base procs. So, as your data set grows, the whole system slows down, and the time to complete the entire thing increases. ... Especially if it takes 4 hours one way and *40* hours your way.

True, but you misunderstand my macros. I do use PROC IML in them but I use it for very low-RAM-consuming computation. In fact, the PROC IML there uses the output dataset of PROC REG, which would be a constant matrix, typically a 17-variable by 6-observation dataset, for either 25-case dataset or 250,000-case dataset. In other words, if it took 4 hours for PROC REG, it'd also take 4 hours for my macros too.

> A really important aspect here is the choice of the proc. ... For example, evaluation of higher-order interactions needs to be done in conjunction with appropriate model diagnostics, including diagnostic plots, so someone can tell whether the errors are normal, the residuals are independent, the errors are identically distributed, the data lack outliers and/or leverage points, ...

True, and these are what I agree ODS Graphics is good at.

Best, Yung-jui


Back to: Top of message | Previous page | Main SAS-L page