Proc Expand Puzzle

[genemroz]

Esteemed advisers:

I’ve been experimenting with Proc Expand.  Specifically, with the transformation MOVTVALUE.  The documentation says: “They can be viewed as combinations of the moving average (CUAVE, MOVAVE, CMOVAVE) and the moving standard deviation (CUSTD, MOVSTD, CMOVSTD), respectively”.  So I was expecting that for a normal distribution, MOVTVALUE would cluster around a value of 0 regardless of the mean of the distribution.  

But that’s not what I found. See code below where I created two normal distributions with a mean of 0 and 1.  I then used Proc Expand to compute MOVTVALUE and then plotted the result and computed the mean of the t-values.  

The plot of t-values for a normal distrubution with a mean of 0 looks as I would expect with values clusters around of mean of (near) zero.  But the plot for t-values for normal distribution of a mean of 1 looks very different. 

What am I misunderstanding?  Thanks in advance for any insights you can provide.

data have;
call streaminit(123);
do i=1 to 100;
test0=rand('normal',0);
test1=rand('normal',1);
output;
end;
run;

proc expand data=have out=out method=none;
id i;
convert test0=tvalue0/transout=(movtvalue 5);
convert test1=tvalue1/transout=(movtvalue 5);
run;

proc sgplot data=out;
   series x=i y=tvalue0  /
   name='tvalue0'   legendlabel="tvalue0" markers markerattrs=(symbol=circlefilled);
      series x=i y=tvalue1  /
   name='tvalue1'   legendlabel="tvalue1" markers markerattrs=(symbol=circlefilled);
   xaxis grid;
   yaxis grid label='t-value';
run;

proc means data=out mean;
var tvalue0 tvalue1;
run;

[sbxkoenk · Accepted solution]

What you see looks correct to me:

• PROBLEM: What exactly is the transformation operation MOVTVALUE calculating (on the EXPAND procedure)?
• RESOLUTION:
  t-statistic (t-value) based on moving windows.  t-statistic corresponds to the test statistic of testing if mean being 0 or not. (see TTEST procedure).

For a mean that is not zero you see higher t-values of course. t-values will often become significant then (rejecting null-hypothesis of mean=0).

Koen

[genemroz]

Thanks for the prompt and clear response.  I see where I went off the rails.

Gene