Proc Rank with Weighted Variables
in [SAS]
|
From: my0402 on 20 Feb 2007 10:58 >From an old thread, circa 2005: toby dunn <tobyd...(a)hotmail.com> wrote: > I am trying to create deciles but I need the obs weighted by another > variable. Does Proc Rank have a wieght statement like proc freq does or is > there a better way to get to where I want to go here. PROC UNIVARIATE has a WEIGHT statement, and will compute weighted quantiles for you. Try that. You'll have to specify your deciles, but the proc will allow that. Like this: proc univariate data=meshugineh noprint; var whatever; weight ohsure; output out=YourTrafe pctlpts=10 to 90 by 10 pctlpre=Dec_ ; run; Of course, if the data come from a survey sample with weights - and possibly other design features - then these point estimates will be okay but the standard errors and variances from PROC UNIVARIATE will all be wrong. _caveat_utor_ . HTH, David -- David Cassell, CSC Cassell.Da...(a)epa.gov Senior computing specialist mathematical statistician ================================ If a record's predicted value is in the top 10%, the record falls into a segment_a, else it doesn't. There's also a 2nd algorithm predicting a different segment_b with the same data, also being assigned by the top 10% of that predicted value. A record can't be in 2 segments, so if it does fall into both, it gets assigned to segment_a. The data are weighted. I don't think I'm handling the weighted aspect correctly. data prob; set data_adjust; pred = ( AGE * -.12345 )+ ( EDUC * .67891 )+ ( HHINC * .000246801 )+ ( RACEO * -.098765 )+ etc. + -.0010203 ; run; proc univariate data=prob; var pred; weight pop_wt; output out = pred P90 = pred_90 ; run; proc sql; select pred_90 into :p90 from pred ; quit; data prob; set prob; if pred ge &p90 then SEGMENT = 1; else SEGMENT = 0; run; proc rank data=prob out=prob_rank descending; var pred; ranks pred_RANK; run; Is this correct? Thanks. Myra
From: BruceBrad on 20 Feb 2007 23:13 On Feb 21, 2:58 am, my0...(a)gmail.com wrote: > >From an old thread, circa 2005: > toby dunn <tobyd...(a)hotmail.com> wrote: > > I am trying to create deciles but I need the obs weighted by another > > variable. Does Proc Rank have a wieght statement like proc freq does > or is > > there a better way to get to where I want to go here. > > PROC UNIVARIATE has a WEIGHT statement, and will compute weighted > quantiles for you. Try that. You'll have to specify your deciles, > but the proc will allow that. Like this: > > proc univariate data=meshugineh noprint; > var whatever; > weight ohsure; > output out=YourTrafe > pctlpts=10 to 90 by 10 > pctlpre=Dec_ ; > run; > > Of course, if the data come from a survey sample with weights - and > possibly other design features - then these point estimates will be > okay but the standard errors and variances from PROC UNIVARIATE will > all be wrong. _caveat_utor_ . > > HTH, > David > -- > David Cassell, CSC > Cassell.Da...(a)epa.gov > Senior computing specialist > mathematical statistician > > ================================ > > If a record's predicted value is in the top 10%, the record falls into > a segment_a, else it doesn't. There's also a 2nd algorithm predicting > a different segment_b with the same data, also being assigned by the > top 10% of that predicted value. A record can't be in 2 segments, so > if it does fall into both, it gets assigned to segment_a. > > The data are weighted. I don't think I'm handling the weighted aspect > correctly. > > data prob; > set data_adjust; > pred = > ( AGE * -.12345 )+ > ( EDUC * .67891 )+ > ( HHINC * .000246801 )+ > ( RACEO * -.098765 )+ > etc. + > -.0010203 > ; > run; > > proc univariate data=prob; > var pred; > weight pop_wt; > output > out = pred > P90 = pred_90 > ; > run; > > proc sql; > select > pred_90 > into > :p90 > from pred > ; > quit; > > data prob; > set prob; > if pred ge &p90 then SEGMENT = 1; > else SEGMENT = 0; > run; > > proc rank data=prob out=prob_rank descending; > var pred; > ranks pred_RANK; > run; > > Is this correct? > > Thanks. > Myra
From: BruceBrad on 20 Feb 2007 23:16 This makes sense up to the proc rank statement. Why are you including this? Haven't you solved your problem in the previous data step?
From: my0402 on 21 Feb 2007 10:33 On Feb 20, 11:16 pm, "BruceBrad" <BruceB...(a)iname.com> wrote: > This makes sense up to the proc rank statement. Why are you including > this? Haven't you solved your problem in the previous data step? Though segment_a is the dominant if the predicted value falls in both, another solution is taking the segment in which the weighted record has the higher rank. That's why I want to look at weighted ranks. Sorry I didn't make that clearer to begin with. Myra
From: my0402 on 21 Feb 2007 10:34 On Feb 20, 10:58 am, my0...(a)gmail.com wrote: > >From an old thread, circa 2005: > toby dunn <tobyd...(a)hotmail.com> wrote: > > I am trying to create deciles but I need the obs weighted by another > > variable. Does Proc Rank have a wieght statement like proc freq does > or is > > there a better way to get to where I want to go here. > > PROC UNIVARIATE has a WEIGHT statement, and will compute weighted > quantiles for you. Try that. You'll have to specify your deciles, > but the proc will allow that. Like this: > > proc univariate data=meshugineh noprint; > var whatever; > weight ohsure; > output out=YourTrafe > pctlpts=10 to 90 by 10 > pctlpre=Dec_ ; > run; > > Of course, if the data come from a survey sample with weights - and > possibly other design features - then these point estimates will be > okay but the standard errors and variances from PROC UNIVARIATE will > all be wrong. _caveat_utor_ . > > HTH, > David > -- > David Cassell, CSC > Cassell.Da...(a)epa.gov > Senior computing specialist > mathematical statistician > > ================================ > > If a record's predicted value is in the top 10%, the record falls into > a segment_a, else it doesn't. There's also a 2nd algorithm predicting > a different segment_b with the same data, also being assigned by the > top 10% of that predicted value. A record can't be in 2 segments, so > if it does fall into both, it gets assigned to segment_a. > > The data are weighted. I don't think I'm handling the weighted aspect > correctly. > > data prob; > set data_adjust; > pred = > ( AGE * -.12345 )+ > ( EDUC * .67891 )+ > ( HHINC * .000246801 )+ > ( RACEO * -.098765 )+ > etc. + > -.0010203 > ; > run; > > proc univariate data=prob; > var pred; > weight pop_wt; > output > out = pred > P90 = pred_90 > ; > run; > > proc sql; > select > pred_90 > into > :p90 > from pred > ; > quit; > > data prob; > set prob; > if pred ge &p90 then SEGMENT = 1; > else SEGMENT = 0; > run; > > proc rank data=prob out=prob_rank descending; > var pred; > ranks pred_RANK; > run; > > Is this correct? > > Thanks. > Myra
|
Next
|
Last
Pages: 1 2 Prev: Q: resolving Windows environment variables Next: PROC MEANS SAS DOCUMENTATION ERROR ?? |