Overlay Series With Regression/Trend Lines for Different (Half) Series using PROC SGPLOT

[adejames2000]

Hello all,

I am trying to lay a regression line on each of two different (half) series covering different times of period.

proc sql;
    create table have as
        select date
             , region
             , sum(sale) as sale
        from sashelp.pricedata
        group by region, date
        order by region, date
    ;
quit;

data have2;
	set have (rename=(region=period));
	if period=3 then delete;
	if period=2 then monthyear=intnx('year',date,5, "sameday");
		else monthyear=date;
	drop date;
	format monthyear monyy5.;
run;

/*CODE 1 PRODUCES SERIES AS EXPECTED*/
proc sgplot data=have2;
    series x=monthyear y=sale / group=period markers;
    label period="PERIOD";
run;

/*CODE 2 PRODUCES REGRESSION LINE AS EXPECTED*/
proc sgplot data=have2;
    reg x=monthyear y=sale / group=period ;
    label period="PERIOD";
run;

/*CODE 3 DOES NOT PRODUCES SERIES OVERLAID WITH REGRESSION LINE*/
*When I add the regression line, all the series line are all over the place;
proc sgplot data=have2;
    series x=monthyear y=sale / group=period markers;* markerattrs=(color=lightgreen size=10 symbol=trianglefilled);
	 reg x=monthyear y=sale / group=period ;
    label period="PERIOD";
run;

I saw a similar example here but it doesn't work for me, presumably because the periods do not overlap.

Below is the plot obtained from CODE 1, I just need to have the regression lines through EACH (half) series without distorting the series as CODE 3 does.

Please help, thank you!

[ballardw · Accepted solution]

Try forcing a sort prior to the last Sgplot and see if that helps. It did on my system (Thank you for using a set we have access to).

I an guessing that without the specified sort order that mixing Reg with series brings up some internal question about "order", which is typically the cause of that type of connections in a series plot.

proc sort data=have2;
   by period monthyear;
run;

proc sgplot data=have2;
    series x=monthyear y=sale / group=period markers;* markerattrs=(color=lightgreen size=10 symbol=trianglefilled);
	 reg x=monthyear y=sale / group=period ;
    label period="PERIOD";
run;

[adejames2000]

I previously only sorted   "by monthyear", sorting by "by period monthyear" does the trick. Thank you very much.