At and above a specific cutoff value, sensitivity is the percentage of those with the outcome of interest that are detected using your logistic model:  the percentage of customers who are more likely to respond.  Sensitivity is the percentage of those without the outcome of interest that are detected using the model:  the percentage of customers who are not likely to response.  The percentage of false positives is the percentage of customers that your model predicts as more likely to respond who in fact do not respond.  The percentage of false negatives is the percentage of customers that your model predicts as not likely to respond who do in fact respond. In your example at a cutoff of 0.20 or more, your model picks up only 16.9% [=sensitivity] of customers who are more likely to respond, and 3.7% [100% - 96.3% (=specificity)] of customers who are not likely to respond.  However, 75.3% [=% of false positives] of those your model predicts as likely to respond will in fact not respond, though 94.2% [=100% - 5.8% (% of false negatives)] of those your model predicts as not likely to respond will in fact not respond. You can visualize this better in a two-by-two table like the following: Test prediction      % likely to respond  % not likely to respond  Total   0.20 or more              9,516                  29,047          38,563   < 0.20                  46,745                753,000          799,745 Total                    56,261                782,047          838,308       Sensitivity =  9,516 /  56,261 = 16.9%       Specificity = 753,000 / 782,047 = 96.3%   False positives =  29,047 /  38,563 = 75.3%   False negatives =  46,745 / 799,745 =  5.8%  To pick an appropriate test prediction cutoff, you have to balance the costs vs. the benefits. Using the % false positives as one criterion, of every four customers you tried to contact, on average only one of them would be likely to respond.  If contacting custormers is relatively cheap, you might not worry so much about this false positive % and prefer to increase the percentage of those likely to respond who in fact do respond (that is, to increase the sensitivity).  At a test prediction cutoff of 0.05 or above, only one of nine customers you tried to contact would be likely to respond [=100%-89% false positives] but you would in fact be able to detect more than three-quarters of those customers that would likely respond [sensitivity=77.9%]. ... View more

Does the first, "hard-coded" version work in your bigger problem? If not, perhaps the problem is that the CALL SYMPUT statements do NOT work to create the macro variables, &VOLUME and &PRICE, in the same PROC IML step.  These statements would not work in a DATA step.  You may have to explain more details about the bigger problem you are trying to solve.  Or, submit this problem to the separate community SAS thread on PROC IML. ... View more

Why are you creating these macro variables in the first place?  If you are trying to extract the columns from the matrix, XXX, to create the matrix, VWAP, why not simply do the following?      VWAP=XXX[,2:3]; Or, if you also need the matrices, VOLUME and PRICE, why not do the following?       VOLUME=XXX[,2];       PRICE=XXX[,3]; ... View more

Since the AUC for your model exceeds 0.50, your model predicts the outcome better than chance.  However, you might also specify the PROC LOGISTIC MODEL statement option, CTABLE, to print a classification table of your outcomes including the sensitivity, and the specificity and plot the receiver-operatiing characteristic curve of your model (cf., the documentation). Then you can consider the trade-offs between higher sensitivity and lower specificity (finding most of the customers likely to purchase tennis tickets but also. finding many who are unlikely to purchase these tickets) and between lower sensitivity an higher specificity (finding fewer customers likely to purchase tennis tickets but also eliminating many of those who are unlikely to purchase these tickets). ... View more

Try the following code.  It's more complicated than usual because your variable, MONTH, is a character variable that may not be compatible with a variable specified in the PLOT statement of PROC GPLOT. Another alternative would be a linear regression using PROC REG, regressing income on month, and looking for positive, neutral, or negative regression coefficients for month. ===================================================================================== * Formats for month variables; proc format;    value $monthfm "January"="1" "Februrary"="2" . . .  "December"="12";    value month 1="January" 2="February" . . . 12="December"; run; * Convert month-names to numbers from 1 to 12; data new;     set;     monthnum=input(put(month,$monthfm.),8.0);     output new; run; * Sort data by name and number of month; proc sort data=new equals;    by name monthnum; run; * Plot income vs. month and interpolate a linear regression line; *  for each name; symbol1 v=plus i=rl; proc gplot data=new;    by name;    plot income*monthnum;    format monthnum month.; run; quit; ... View more

Can you take multiple random samples of your data, run PROC MIXED on each of these samples, and then summarize the results across the samples?  If your sample were one-half the size of your original data set, the corresponding matrices would be only one-fourth the size as those from your original data set. Or can you group observations near to one another and summarize/average their yields to reduce your sample size? ... View more

Whether or not the ROC curve is based on a single measurement or repeated measurements, the choice of a cutoff depends on the trade-off between sensitivity and specificity.  Ask the doctor whether sensitivity is more important than specificity or vice-versa and how much more important one is than another.  For example, the sensitivity at or above level A of TNM if the proportion of all true pregnancies that are detected; the specificity below level A of TNM is the proportion of those who are truly not pregnant who are declared as not pregnant. Or, ask the doctor whether level A of TNM yields an acceptable positive predictive value of the test:  What proportion of the tests positive for pregnancy are associated with actual pregnancies?  Unlike sensitivity and specificity, predictive values depend on the prevalence of the outcome (=pregnancy) in the population being tested; if pregnancy is rare in this population, positive predictive values are lower so that many positive tests are detected among those who are NOT pregnant. I can't answer your other questions about differences in estimates of the area-under-the-curve between PROC LOGISTIC and PROC GLIMMIX.  In general, multiple tests showing consistent results should improve sensitivity and specificity.  Since you have no time indexes on your tests, you probably should consider either compound-symmetry or unrestricted variance-covariance matrices that do not depend on time. ... View more

Create a new format value for the CLASS variable you want to specify a reference level for.  For example, if "Barley" is the commodity you want as your reference level for all your commodities, use a PROC FORMAT VALUE statement to create a formatted value that sorts at the end of all formatted values by preceding that value with one or more right braces ("}"):     proc format;        value $comm "Barley"="}} Barley";    run;    proc quantreg . . . order=formatted;        class commodity;        . . . .        format commodity $comm.;    run;         This syntax will make "Barley" the reference level for the classification variable, commodity. ... View more

The STRATA statement in PROC LOGISTIC is used to define variables that identify matched sets of observations so that these matched sets can be analyzed using conditional logistic regression, not the usual unconditional logistic regression. Given your description, include your "stratum" variable in the CLASS statement (specifying a reference level) and in the MODEL statement as an independent variable.  Then specify your ordinal variable as the dependent variable in the MODEL statement, which PROC LOGISTIC will interpret as an ordinal variable under the default LINK option specification. ... View more

1) PROC DISCRIM parameterizes the discriminant functions differently from other programs.  You can estimate the posterior probability of any given observation being in a particular "dependent variable" category using the discriminant function for that category and the prior probability of that category. 2) As the documentation states, PROC DISCRIM places each observation in the category from which it has the smallest generalized square distance based on the different discriminant functions.  This distance is the squared Mahalanobis distance between the observation and the mean values of each category using the pooled covariance matrix and the prior probabilities for each category (in the fish example).  The discriminant scores are -0.5*this distance for each category.  For each observation, the OUT= data set includes these posterior probabilities by category, and the OUTD= data set includes the category-specific density estimates.  You can also specify the PROC DISCRIM statement option, SCORES, to add the discriminant function scores for each categoy to the OUT= data set. 3) Yes, you are right.  The rows in Figure 31.5 are the true species of the observation and the columns the species that the observation that the discriminant function classifies the observation into.  4) The CROSSVALIDATE option uses the other N-1 observations to classify any one observation.  You can use the K=n or the KPROP=p options of the PROC DISCRIM statement to use only the observation's K-th nearest neighbors or the K-th proportion of observations to classify an observation into specific groups.  I do not think that you can use a random sample of K observations to classify a specific observation. 5) Figure 31.1 describes the number of observations used, the actual probability of each category, and the default prior probability of each category in the discriminant analysis.  The PRIORS statement allows you to change the prior probabilities from their default of being equal (that is, independent of the sample size in the categories).  Changing this statement would change the estimated discriminant functions.  Figure 31.2 shows the default information for the pooled covariance matrix, which is the basis of the measure of the measure of squared distance; you can change this basis to the individual within-category covariance matrices to calculate these distances (option, POOL=NO).   ... View more