The dates aren't extremely important.  When I used your macro with a small test sample I removed the date field.  It worked great, the date would be ideal to pull the acct number with the lowest date.    The numbers aren't sequential either, which makes another problem.  So the min(account) doesn't necessarily have to be the earliest one issued.  I'm not sure how to QC this after the fact.  I have my own solution that isn't great and very slow, I'm taking known accounts that are good examples:  Say someone was acquired from a different company, so they have a reference number that ties it to the account but was never a real account number, then loses their card every weekend at a bar, so they have 20 different card numbers.    Using an example like that I can track down all the individual cross reference numbers manually and see if my process works, but that doesn't lend a great deal of confidence to the population.  ... View more

If there isn't a value in the format it keeps the value, in this case it also happens to be a 1.    If you add an OTHER to your format it will resolve as expected. This is VALUE, not INVALUE.    proc format ; value product 2, 3='1' other='0'; run; data have; infile cards; input test; cards; 1 2 3 ; run; data want; set have; where put(test, product.)='1'; format test product.; run; proc print data=want; run; ... View more

Hi Patrick, Thank you very much.   Mirisa ... View more

I'm getting the following error:   ERROR: Type mismatch for key variable ACCTNBR at line 43 column 19. ERROR: Keys and data must be specified in the same order and have the same types as given in DefineKey/DefineData at line 43 column        19.   I'm not sure what to do with this.  A quick google search for the error hasn't helped ... View more

@Steelers_In_DC wrote: I don't know what you mean by 'between'.  I've never seen that.  I'm not familiar wiith indsname either, when I run that I get:   does not contain any members INDSNAME is an option that creates a variable that can store the name of the dataset a record comes from, when appending multiple datasets together.    Post your log. As long as you're on SAS 9.3+ this should work, if you're on 9.2 then it won't. ... View more

Glad I was able to help. I forgot to mention that any actions taken to benefit from your model, like preventing churn or default, need to be effective - an attractive customer proposition. This can be tracked by looking at customer behaviour following the campaign. A poor campaign can be worse than doing nothing at all.   An example of an attractive customer retention offer would be to give a significant discount and/or lock in a fixed price if the customer was prepared to commit to a 2 year contract instead of just staying month-by-month. ... View more

I've only ever used it to truncate what I needed, never thought of it to force a number longer.  I guess it's nothing I've ever come across.  Very useful.  A nice tool to add to the toolbox.    Thanks!! ... View more

You also mentioned about the height of the bars not corresponding to the height of the labels. This is because you're using a stacked bar chart - the labels are for the value inside a single bar segment, whereas the numbers along the axis are for the combined/stacked bars.   I think it might be better to use a paired/grouped bar chart, rather than a stacked bar chart for this data, and here is the code for that:   data graph_prep; length pass_fail $4; format date date9.; input date date9. pass_fail count; datalines; 31may2016 Fail 101 31may2016 Pass 358 30jun2016 Fail 101 30jun2016 Pass 358 31jul2016 Fail 97 31jul2016 Pass 362 ; run;   axis1 label=none value=none; axis2 label=none; axis3 minor=none; legend1 label=none; proc gchart data=graph_prep; vbar pass_fail / group=date space=0 type=sum sumvar=count noframe inside=sum maxis=axis1 gaxis=axis2 raxis=axis3 subgroup=pass_fail legend=legend1; run;   ... View more

Ah, if it was a snake it would have bit me.  Thanks! ... View more