About StatDave

StatDave · ‎11-22-2023

See the "Statistical Background in Linear Regression" section in the Introduction to Regression Procedures chapter of the SAS/STAT User's Guide. Formulas are shown in the subsections.

StatDave · ‎11-21-2023

You appear to be using JMP. You should post your question in the JMP Communities, not the SAS Communities.

StatDave · ‎11-19-2023

Yes, using the log link you model the log of the event probability and then exponentiation of parameters can be used to estimate relative risks. But using the log link can result in estimation problems because the log link does not keep predicted values restricted to the valid probability range. However, you can also estimate relative risks using a logistic model followed by use of the NLMeans macro. These and other approaches are covered in this note.

StatDave · ‎11-15-2023

The Generalized Estimating Equations (GEE) model is a population-averaged marginal model and supports generalized linear models with various links for clustered data. The procedure to use is PROC GEE. Specify the response distribution in the DIST= option in the MODEL statement. Specify the log link with LINK=LOG option in the MODEL statement. Add the REPEATED statement with the SUBJECT= option to specify the variable whose levels indicate the clusters of correlated observations. See the examples in the PROC GEE documentation.

StatDave · ‎11-15-2023

That is because the parameter estimate on VAR1 is positive in this example, negative in the previous one.

StatDave · ‎11-14-2023

The value of your predictor (VAR1) that corresponds to the optimal threshold can be displayed by specifying VAR1 in the ID statement and using ID=ID in ROCOPTIONS: proc logistic data=tab rocoptions(optimal=youden id=id); id var1; model resp_q(event='Répondeur')=var1 / outroc=roc_var1 ; run; If you do that, you will see in the ROC plot that the optimal threshold corresponds to VAR1=14. You can use that to make a variable of predicted response levels and produce the 2x2, predicted by actual table: data x; set tab; pred=(var1<=14); run; proc freq data=x; table pred*resp_q; run; The resulting table agrees with the _POS_, _NEG_, _FALPOS_, and _FALNEG_ values in the OUTROC= table.

StatDave · ‎11-12-2023

You cannot obtain odds ratios for all levels of a categorical (CLASS) variable. An odds ratio compares the event odds of two levels of a categorical variable. So, for a categorical with two levels, there can only be one odds ratio. The plot you show (which is of hazard ratios, not odds ratios) apparently compares levels of some other binary variable (with placebo and MTX levels) within each gender rather than comparing the response hazards of the two genders. That is how there can be hazard ratios for both genders. You can use the ESTIMATE statement to estimate odds ratios and then plot them using PROC SGPLOT. PROC GEE is a newer procedure specifically for fitting the GEE model and is the recommended procedure when fitting that model. Use the EXP and CL options in the ESTIMATE to obtain the odds ratio estimates and confidence limits. Using the wheezing data in the Getting Started section of the GENMOD documentation, the following fits the GEE model, estimates and saves the odds ratios, and plots them. The coefficients for CITY compare the odds of its two levels and exponentiate that difference. For a continuous variable, the 1 coefficient selects its single parameter that is exponentiated to obtain the odds ratio for the difference of one unit. proc gee data=six; class case City; model wheeze(event='1') = City Age Smoke / dist=bin link=logit; repeated subject=case / type=exch; estimate 'city' city 1 -1, 'age' age 1, 'smoke' smoke 1 / exp cl; ods output estimates=or; run; proc sgplot noautolegend; highlow high=upperexp low=lowerexp y=label / highcap=serif lowcap=serif; scatter y=label x=expestimate; refline 0 / axis=x; yaxis label=' '; xaxis label='Odds Ratio'; run;

StatDave · ‎11-11-2023

_GROUP should be there just as in this example: proc surveyreg data=sashelp.cars; class origin; effect s=spline(enginesize/naturalcubic); model mpg_city=s origin horsepower; store sr; run; proc plm restore=sr; EFFECTPLOT sliceFIT (X=enginesize sliceby=origin)/ CLM; ODS OUTPUT sliceFITPLOT = fp; run; PROC SGPLOT DATA=fp NOAUTOLEGEND; BAND UPPER = _UCLM LOWER = _LCLM X=_XCONT1/TRANSPARENCY=.3 group=_group; SERIES Y = _PREDICTED X=_XCONT1/group=_group; run;

StatDave · ‎11-11-2023

The PLM plot that you show is a SLICEFIT plot, but your code does not request that and other items in the plot are not compatible with your code. However, since the plot you show is a SLICEFIT type of plot with SLICEBY=SEX, you need to specify GROUP=_GROUP in both your BAND and in your SERIES statements.

StatDave · ‎11-10-2023

If you are only interested in testing for a linear association between the two continuous variable (on the percent or proportion scale), then yes, you could use the Pearson or Spearman correlation in PROC CORR to get that test. Do not categorize either variable as that just wastes information. If you think the association might be important, but not necessarily linear, then you could fit a fractional logit model as described in this note. With that model you could even use a spline on the predictor (accuracy measure) to allow for an indeterminate shape on the association. That is done by specifying the SPLINE effect in the EFFECT statement in PROC GLIMMIX or PROC LOGISTIC.

StatDave · ‎11-08-2023

You would use the actual proportions as your predictor values.

StatDave · ‎11-05-2023

The offset, log of population in your case, is just another predictor in the model. The only difference is that it is restricted to have a parameter estimate equal to 1. In the estimation process, all parameter estimates are adjusted for the presence of the others. Since the response function that you are modeling is the log of the Poisson mean, that is what the parameter estimates apply to. So, the X1 parameter is the effect of a unit increase in X1 on the log Poisson mean - the log mean count. However, computing a linear combination of the parameters, without involving the offset value, predicts the rate, not the mean. See the discussion of all this in this note. BTW, you should never specify a variable in the CLASS statement that is not used elsewhere in the model specification. Missing values in such variables will cause observations to be ignored which don't need to be. Since you are not using COUNTY in the model and it is the only thing in the CLASS statement, you should drop the CLASS statement.

StatDave · ‎11-03-2023

No, I'm referring to an ordinary logistic regression model fit to binary response data: log(p/(1-p)) = Xβ, where p=Pr(Y=1), as can be fit by PROC LOGISTIC or, since this is repeated measures analysis, by PROC GEE or PROC GLIMMIX - that is, a generalized linear model with binomial distribution and logit link function. Examples can be found in the documentation of all of those procedures.

StatDave · ‎11-03-2023

If you think of the classic S shape of a logistic curve, the lower half of it probably is quite similar to what you see in your data. So, a logistic model might fit well. Since the data consist of repeated measures, you could use either a Generalized Estimating Equations model with PROC GEE or a random effects model with PROC GLIMMIX, depending on whether you want a population-averaged model or a subject-specific model. In the model you could include both time and your iron measures as predictors.

StatDave · ‎10-25-2023

GLMSELECT fits the "general linear model" that assumes that the response distribution is normal and it directly models the response mean. GENMOD fits the "generalized linear model" which allows for any response distribution in a family of distributions and it models a function (the "link" function) of the response mean. If you don't use model selection in GLMSELECT, it will fit the same model as GENMOD when you select the normal response distribution and the identity link function, though different estimation methods are used. The estimated parameters (some call them "weights") of the model can be found in the "Parameter Estimates" table from GLMSELECT. The model is then mean(y)=b0+b1*x1+ ... similar to the quoted text you show.

Online Status	Offline
Date Last Visited	4 weeks ago

Re: Association between nominal variable and ordinal variable

Re: Association between nominal variable and ordinal variable

Re: Association between nominal variable and ordinal variable

Re: Association between nominal variable and ordinal variable

Re: Odds ratio and relative risk from procedure causaltrt

Re: Confidence Interval for Percent Increase

Re: Confidence Interval for Percent Increase

Re: Confidence Interval for Percent Increase

Re: Undersampling in PROC PROC HPSPLIT / Adjusting for prior probablit...

Re: Appropriate Analysis for Score (Ordinal) Data - with example data

Re: Model for Correlated data

Re: Association between nominal variable and ordinal variable

Re: Association between nominal variable and ordinal variable

Re: Association between nominal variable and ordinal variable

Re: Association between nominal variable and ordinal variable

Re: Odds ratio and relative risk from procedure causaltrt

Re: Confidence interval for the regression fit

Re: Interpretation of Dummy Variables in Multiple Regression with {0-1...

Re: PROC GENMOD link option values for binomial distribution

Re: population-average marginal generalized linear models with logarit...

Re: Proc logistic rocoptions: problem with contingency table

Re: Proc logistic rocoptions: problem with contingency table

Re: How to create an odds ratio figure with hierarchical regression

Re: Editing slicefit type of effectplot (in proc plm) through proc sgp...

Re: Editing slicefit type of effectplot (in proc plm) through proc sgp...

Re: What test to use for checking the association between a percentile...

Re: event/trials predictors and outcomes in glimmix

Re: offset interpretation for proc genmode

Re: fitting an exponential decay model to binary data (repeated measur...

Re: fitting an exponential decay model to binary data (repeated measur...

Re: What the equation for the model fitted by proc glmselect