Greetings good SAS people,

On July 16 ff. in "Use SAS to fit this curve?", I inquired as to what tool
I could use to find a best fit for some data.

In subsequent manual tinkering, I have found that this data (appended) is
very nicely fit by a two-term exponential equation such as:

   Sites  =  773e^(-0.27*Chems)  +  156e^(-0.067*Chems)

(It is plotting the number of major U.S. hazardous waste sites at which
e.g. for Chems=1, 1 or more chemicals were in a pathway to which people
were exposed. Overall, the curve shows how often (the number of waste
sites) there is a high number of chemicals in the pathway, when folks are
exposed.)

The data itself makes a very "nice" curve to the naked eye, but a one-term
exponential simply wasn't cutting it. Two exp. terms were found to do a
lovely fit. However, this was all done "by hand" (in Excel, watching how
well my model fit vs. the data on a graph) and I have no tool to give me
feedback relative to R^2 values etc. (I guess I should bone up and can
probably do regression in Excel?... anybody have a quick webpage reference
handy?) Nor do I know whether/if SAS (or some other s/w) has the ability to
make and/or fit a two term exponential. I played around with Insight as a
result of the previous post, but it has so many arcanes variables (a.k.a.
I'm such a newbie) I can't tell whether it can try to fit with a two term
exponential. (I'm not interested in polynomial fits, and that's the only
thing I was able to figure out how to get Insight to work.)

Thanks in advance if you can help!!

   Mike, Atlanta

***

input Chems Sites;

cards;

 1 743
 2 588
 3 475
 4 390
 5 321
 6 266
 7 214
 8 185
 9 161
 10 136
 11 111
 12 102
 13 87
 14 74
 15 69
 16 61
 17 56
 18 52
 19 47
 20 46
 21 44
 22 39
 23 36
 24 33
 25 29
 26 26
 28 24
 29 22
 30 20
 31 19
 32 18
 37 16
 38 15
 39 13
 40 11
 41 10
 42 9
 48 7
 49 6
 51 5
 56 4
 58 3
 62 2

;

PROC CAPABILITY, part of the SAS/QC package, has some very nice tools for
curve-fitting. Part of PROC CAPABILITY's histogram functionality was added
to PROC UNIVARIATE in Base SAS, but not the curve-fitting features. You do
have to tell it what distribution to attempt to fit.

Jonathan Siegel


On Fri, 1 Aug 2003 11:04:00 -0400, Mike Fay <rmf4@CDC.GOV> wrote:

>Greetings good SAS people,
>
>On July 16 ff. in "Use SAS to fit this curve?", I inquired as to what tool
>I could use to find a best fit for some data.
>
>In subsequent manual tinkering, I have found that this data (appended) is
>very nicely fit by a two-term exponential equation such as:
>
>   Sites  =  773e^(-0.27*Chems)  +  156e^(-0.067*Chems)
>
>(It is plotting the number of major U.S. hazardous waste sites at which
>e.g. for Chems=1, 1 or more chemicals were in a pathway to which people
>were exposed. Overall, the curve shows how often (the number of waste
>sites) there is a high number of chemicals in the pathway, when folks are
>exposed.)
>
>The data itself makes a very "nice" curve to the naked eye, but a one-term
>exponential simply wasn't cutting it. Two exp. terms were found to do a
>lovely fit. However, this was all done "by hand" (in Excel, watching how
>well my model fit vs. the data on a graph) and I have no tool to give me
>feedback relative to R^2 values etc. (I guess I should bone up and can
>probably do regression in Excel?... anybody have a quick webpage reference
>handy?) Nor do I know whether/if SAS (or some other s/w) has the ability to
>make and/or fit a two term exponential. I played around with Insight as a
>result of the previous post, but it has so many arcanes variables (a.k.a.
>I'm such a newbie) I can't tell whether it can try to fit with a two term
>exponential. (I'm not interested in polynomial fits, and that's the only
>thing I was able to figure out how to get Insight to work.)
>
>Thanks in advance if you can help!!
>
>   Mike, Atlanta
>
>***
>
>input Chems Sites;
>
>cards;
>
> 1 743
> 2 588
> 3 475
> 4 390
> 5 321
> 6 266
> 7 214
> 8 185
> 9 161
> 10 136
> 11 111
> 12 102
> 13 87
> 14 74
> 15 69
> 16 61
> 17 56
> 18 52
> 19 47
> 20 46
> 21 44
> 22 39
> 23 36
> 24 33
> 25 29
> 26 26
> 28 24
> 29 22
> 30 20
> 31 19
> 32 18
> 37 16
> 38 15
> 39 13
> 40 11
> 41 10
> 42 9
> 48 7
> 49 6
> 51 5
> 56 4
> 58 3
> 62 2
>
>;