Hello @JohnPederson,
Thanks for providing sample data (to be read without the DSD option of the INFILE statement). It turns out that the values of variable pclinton differ between any two months only by a constant for every hour (hour=0, 1, ..., 23). See the output of a step like this:
proc sql;
select a.hour, b.pclinton-a.pclinton as dpclinton
from example(where=(month=1)) a,
example(where=(month=2)) b
where a.hour=b.hour;
quit;
Therefore, in a plot like this
proc sgplot data=example;
series x=hour y=pclinton / group=month;
run;
we see 12 parallel "curves."
The same holds for all eleven other analysis variables phrm, ..., channel6 as well. The variable labels "... Predicted Values" suggest that pclinton, phrm, etc. do not contain measured values, but predictions based on some statistical model, which explains the "systematic" differences described above.
Since the correlation coefficient is invariant under linear transformations f(x)=ax+b with a>0, in particular translations (f(x)=x+c), the correlation between, e.g., pclinton and phrm must be the same for every month: If the ("predicted") values are X0, ..., X23 (for pclinton) and Y0, ..., Y23 (for phrm) for one month, they are X0+c, ..., X23+c and Y0+d, ..., Y23+d for another month, with constants c and d depending only on the month. The same applies to all other pairs of analysis variables (excluding missing values).
The "parallel" results of your linear regressions can be explained similarly.
... View more