Instead of total monthly sales (S) as your dependent variable, you could model average monthly sales per customer (S/C).  You could also examine the independent variables added to the regression on S of C that change the sign of the regression coefficient of C.  What combination of these independent variables change this sign and why?  Perhaps such an examination would identify subgroups based on levels of these independent variables or interactions between C and these independent variables that might be worth further study and modelling. ... View more

Any of the linear model procs should give you what you need.  See the documentation for PROC GLM, PROC MIXED, PROC GENMOD and PROC GLIMMIX (most likely the latter two since I think your data will probably have a non-normal distribution). Steve Denham ... View more

Different shops have often used different products to do their stats. Once ingrained it can be culturally difficult for the organization to change. Program editor choices can be just as bad. I am glad I moved back to a SAS shop as my last job used SPSS. Every year there was an "upgrade" to the software and we usually lost at least one key step in our processes. We had even kept a license for a version that was 4 upgrades out of date because no one could figure out how to duplicate the output a client wanted using the later versions. ... View more

Read about the phenomenon called "suppression".  One article that describes it is the following:    Lynn HS.  Suppression and confounding in action.  The American Statistician 2003 Feb;57(1):58-61. ... View more

thanks for the great answer, pagemiller. ... View more

I hate it when this happens.  Region and marketing campaign are completely confounded in this design.  I do not see how you can attribute any differences found to the effect of the marketing campaign when there are certain to be regional differences.  You may be saved by the time factor, if the measures of sales in each region are pre and post campaign values.  A repeated measures model that looks at the interaction between time and campaign, with region as a random effect might be able to salvage your data. Steve Denham ... View more

Any comments on this paper for my question? http://support.sas.com/resources/papers/proceedings10/213-2010.pdf Second, can we use propensity matching here? Suppose we have 2000 members available for creating a control and test groups. 1. Randomly select one members from them 2. Use propensity score matching to find a matching member 3 Repeat step 1 and 2 until exhaust all members ps. SAS paper for propensity score matching: http://www2.sas.com/proceedings/sugi26/p214-26.pdf ... View more

These two Patterns you are talking about is linear correlation (i.e. Pearson correlated coefficient ) which conform to Normal distribution (i.e. independent for each other). you can use correlation matrix, once you make sure that they are non-correlated (i.e. each airline make its decision is independent ) . proc varclus is implementing multi-variables analysis, which is nothing with linear correlation. he use principal component analysis. The first principal component is a weighted average of the variables that explains as much variance as possible. For example: if A and B belong to a group that means A and B can explain most variance of either the correlation or the covariance matrix . KSharp ... View more

Here is one source: http://www.ers.usda.gov/data-products/international-macroeconomic-data-set.aspx ... View more

Geoffry, Opps... I guess I shouldnt try to do too many things at the same time...the pattern modifier should have been a 'i' instead of 'o'. It should have been: Ind = PrxMatch( '/a-z/i' , Var1 ) ; This will match a-z and A-Z characters as the 'i' pattern modifier tells the RegEx engine to use case insensitivity.  There is no need for the 'o' pattern modifier since the pattern is explicitly and does not change.  Which means SAS by default will only compile it once.  Since Var1 is suppose to only be length of 1 the varibale does not need to be trimmed of any leading or trailing spaces.  If one only wanted to check for lower case letters then: Ind = PrxMatch( '/a-z/' , Var1 ) ; And for only upper case: Ind = PrxMatch( '/A-Z/' , Var1 ) ; Sorry for the fat fingered and lack of proofing on the last post. ... View more

It is true that strength of evidence requirements vary a lot across disciplines.  -  To go beyond regression, you might want to consider the model building techniques provided by data mining tools - PG ... View more