Please help me understand the datalines input in this SAS note

[cjpsas]

Hi all,

I am trying to learn how to estimate the difference in difference in models with covariates, using the example from this SAS note (https://support.sas.com/kb/61/830.html). In the example (pasted below), I understand the input k1-k12 as the # of means estimated by the LSMEANS statement, but don't fully understand how the datalines (1 -1 -1 1 0 0 0 0...etc) represent the three levels of the covariate of interest. I want to make sure I fully understand this format so I can modify the code for datasets with more covariates or covariates with >3 levels. It seems very basic but any explanation would be greatly appreciated!

data difdif;
        input k1-k12;
        set=1;
        datalines;
        1 -1 -1 1   0 0 0 0     0 0 0 0
        0 0 0 0     1 -1 -1 1   0 0 0 0
        0 0 0 0     0 0 0 0     1 -1 -1 1   
        ;
      %NLMeans(instore=log, coef=coeffs, link=logit, contrasts=difdif,
               title=Difference in Difference of Means)

[acordes]

This is about design matrices.
Run proc reg or proc glm with the option print all, use difference statements and study the design matrix.
You'll see which 1 belongs to the intercept, how differences are constructed.
This is subject to your parameterization choice.
Read this blog and things will become clearer.
https://blogs.sas.com/content/iml/2016/02/24/create-a-design-matrix-in-sas.html

[StatDave · Accepted solution]

As described in that note, C has 3 levels (and A and B each have 2) so that the LSMEANS statement that is shown produces 12 estimates. Hence there are 12 values in each data row of the DIFDIF DATA step. The 12 values in a row multiply the mean estimates from the LSMEANS statement, so if you want a difference in difference estimate at each level of C, you need to construct a list of 12 values to compute the difference in difference at each level of C. Because C is first in the specification, C*A*B, in the LSMEANS statement, the first 4 estimates are for the first level of C. So the first non-zero values in the first data row compute the difference in difference for C=1 and the zeros in the final 8 values ignore the C=2 and 3 levels. Similarly, the second data row picks up just the C=2 estimates to compute the difference in difference for C=2. And similarly for the third data row. 

[cjpsas]

Thank you for the clear explanation!