topic Loess Plot in SAS Programming

Loess Plot

jacksonan123 — Fri, 14 Jun 2019 19:12:55 GMT

I have the following code which works well for a needle plot to which I added a Loess to connect certain points. The Loess worked for 4 of the variables but I get an error message stating, "

ERROR: No LOESS fit produced for group 'PAUC03L' because all computed smoothing criterion values are invalid.

ERROR: No LOESS fit produced for group 'PAUC37L' because all computed smoothing criterion values are invalid."

The data is below and I have looked at it and can't figure out why SAS has an issue with the PAUC03L and PAUC37L data.

They both decrease as they should.

title 'Confidence Interval Range Kfast';
proc sgplot data=boot2 ;
  /*highlow x=ratio low=low high=high / type=line;*/
 loess x=ratio y=LOWER_ci/interpolation=cubic group=block;
  highlow x=ratio low=lower_ci high=upper_ci / type=bar group=block   
          lineattrs=(color=black) name='a' attrid=Mono;
  yaxis label='CI range' grid;
  xaxis label='Ratio KaT/KaR' grid values =(0.5 to 2.1 by 0.1); 
      keylegend 'a' / location=inside position=TOP;
  run;

Can some one tell me what is wrong with the boot2 data for PAUC03L and PAUC37L for the Lower_CI?

Group	Block	Upper_CI	Lower_CI	Ratio
1	PAUC03L	84	80	0.62
1	PAUC03L	85	81	0.63
1	PAUC03L	88	85	0.7
1	PAUC03L	100	99	0.8
1	PAUC03U	108	106	1.25
1	PAUC03U	109	106	1.3
1	PAUC03U	113	109	1.5
1	PAUC03U	122	115	2
1	PAUC03U	125	116	2.1
1	PAUC37L	104	102	0.62
1	PAUC37L	104	101	0.63
1	PAUC37L	104	100	0.7
1	PAUC37L	90	88	0.8
1	PAUC37U	98	96	1.25
1	PAUC37U	96	95	1.3
1	PAUC37U	96	93	1.5
1	PAUC37U	94	89	2
1	PAUC37U	94	89	2.1

Re: Loess Plot

PaigeMiller — Fri, 14 Jun 2019 19:15:43 GMT

Just a guess, but maybe you can't do cubic interpolation with just 4 data points.

Re: Loess Plot

jacksonan123 — Fri, 14 Jun 2019 19:57:17 GMT

I tried to play with the options statement and when I do it plots the data
but now gives an error for the previous data. If I use degree 1 which
specifies the degree of the local polynomials to use for each local
regression. 1 specifies a linear fit and 2 specifies a quadratic fit. It
appears that I have a combination of linear for some data and cubic for the
other. The question is how to combine them other wise I may have to use
procgpanel and make separate graphs.

loess x=ratio y=LOWER_ci/interpolation=cubic group=block degree=2;

Re: Loess Plot

Reeza — Fri, 14 Jun 2019 20:31:51 GMT

Is that actually all of the data you have? You're trying to do a 2 degree polynomial regression with 4 points?

Re: Loess Plot

PaigeMiller — Fri, 14 Jun 2019 23:37:08 GMT

I have to admit, I am totally mystified by your comments. Your original code does not use the DEGREE= option, it does use the INTERPOLATION= option, which is what I responded to. And then you ignore the INTERPOLATION option in your reply.

Your original message talks about the lower confidence interval, even though the error messages don't talk about lower confidence intervals at all.

But anyway, the problem, as I implied, is that you have a very small amount of data to do some of this smoothing. Playing with the options won't overcome the fact that you have a very small amount of data.

Re: Loess Plot

jacksonan123 — Sat, 15 Jun 2019 03:01:17 GMT

I was just trying some other options. I would agree with you that a major issue is the amount of data that I have for a complex function.