@jpprovost wrote:
@PaigeMiller wrote:
Well, that's too bad, as I said your specific problem is what PROC SHEWHART does.
In the simplest case, you create limits based upon your 62 observations, and then see if observation 63 is within those limits. The limits are usually computed as (mean ± k*sigma), where k is some constant, perhaps k=3, and sigma is the standard deviation of the 62 observations. So that case can be computed and graphed without PROC SHEWHART. However more advanced cases would be more difficult to program.
I just thought of something. How can I define those limits if my observations are increasing? By example, if I have the market_value as an observation, it will increase from the past 5 years, etc.
No?
One way might be to look at the rate of change. If the value you are concerned with averages a change of 10 per month then you have something to investigate. Such as the count for the current month is less than or equal to the count of the previous month +10 (or some fudge factor). Or if the increase would be better expressed as a percentage then: current count <= pervious month *(1 + percent change).
For example with something like this (simulating some data)
data example;
count=100;
do month= 1 to 60;
m2m = 10 + rand('integer',1,4);
count= count + m2m;
output;
end;
run;
proc means data=example;
var m2m;
run;
You will likely get mean value somewhere between 12.3 and 12.7 for the mean change from month to month, the m2m variable. And a standard deviation about 1.2 so the mean plus or minus 2 standard deviations are likely to be "reasonable" bounds for the current month change from last month.
If you plot your month to month change and see a notable pattern then you might want to consider an adjustment, possibly based on a regression of the curve.
