1          OPTIONS NONOTES NOSTIMER NOSOURCE NOSYNTAXCHECK;
 55         
 56         /* MathLogReg4.sas  */
 57         %include '/folders/myfolders/441s16/Lecture/readmath2b.sas';
 NOTE: Format YNFMT is already on the library WORK.FORMATS.
 NOTE: Format YNFMT has been output.
 NOTE: Format CRSFMT is already on the library WORK.FORMATS.
 NOTE: Format CRSFMT has been output.
 NOTE: Format NFMT is already on the library WORK.FORMATS.
 NOTE: Format NFMT has been output.
 
 NOTE: PROCEDURE FORMAT used (Total process time):
       real time           0.00 seconds
       cpu time            0.00 seconds
       
 
 166        /* Creates data table mathex */
 167        title2 'Logistic regression with more than 2 resp. categories';
 168        /*************** Data step continues ************************/
 169        hsutil = hsgpa+hscalc+hsengl;
 170        if hsutil = . then hsmiss=1; else hsmiss=0;
 171        label hsmiss = 'Missing Any High School Data'; format hsmiss ynfmt.;
 172        
 173        if (0<=mark<=49) then outcome = 'Fail';
 174           else if (50<=mark<=100) then outcome = 'Pass';
 175           else outcome = 'Gone';
 176        /*************************************************************/
 177        
 
 NOTE: The infile '/folders/myfolders/exploremath.data.txt' is:
       Filename=/folders/myfolders/exploremath.data.txt,
       Owner Name=root,Group Name=vboxsf,
       Access Permission=-rwxrwx---,
       Last Modified=18Jan2016:18:34:49,
       File Size (bytes)=44583
 
 NOTE: 579 records were read from the infile '/folders/myfolders/exploremath.data.txt'.
       The minimum record length was 75.
       The maximum record length was 75.
 NOTE: Missing values were generated as a result of performing an operation on missing values.
       Each place is given by: (Number of times) at (Line):(Column).
       99 at 80:24     99 at 117:13    142 at 169:15   2 at 169:22     
 NOTE: The data set WORK.MATHEX has 579 observations and 37 variables.
 NOTE: DATA statement used (Total process time):
       real time           0.01 seconds
       cpu time            0.01 seconds
       
 178        proc freq;
 
 179             tables outcome*passed / norow nocol nopercent missing;
 180        
 
 NOTE: There were 579 observations read from the data set WORK.MATHEX.
 NOTE: PROCEDURE FREQ used (Total process time):
       real time           0.05 seconds
       cpu time            0.05 seconds
       
 181        proc freq;
 
 182             title3 'One at a time cat IVs with proc freq';
 183             tables (course2 sex ethnic tongue hsmiss) * outcome
 184                    / nocol nopercent chisq;
 185        
 186        /* Multinomial Logit model is
 187        
 188             ln(pi1/pi3) = beta01 + beta11 x             Fail vs. Pass
 189             ln(pi2/pi3) = beta02 + beta12 x             Gone vs. Pass
 190        */
 191        
 
 NOTE: There were 579 observations read from the data set WORK.MATHEX.
 NOTE: PROCEDURE FREQ used (Total process time):
       real time           0.18 seconds
       cpu time            0.18 seconds
       
 192        proc logistic data=mathex outest=ParmNames;
 
 193             title3 'Multinomial logit model with proc logistic';
 194             model outcome (ref='Pass') = hsmiss / link = glogit;
 195             contrast 'HS Missing method 1' hsmiss 1;
 196        
 197        /* Find out the parameter names. */
 198        
 
 NOTE: PROC LOGISTIC is fitting the generalized logit model. The logits modeled contrast each response category against the 
       reference category (outcome='Pass').
 NOTE: Convergence criterion (GCONV=1E-8) satisfied.
 NOTE: There were 579 observations read from the data set WORK.MATHEX.
 NOTE: The data set WORK.PARMNAMES has 1 observations and 10 variables.
 NOTE: PROCEDURE LOGISTIC used (Total process time):
       real time           0.10 seconds
       cpu time            0.09 seconds
       
 199        proc transpose data=ParmNames;
 
 200                 run;
 
 NOTE: There were 1 observations read from the data set WORK.PARMNAMES.
 NOTE: The data set WORK.DATA6 has 5 observations and 3 variables.
 NOTE: PROCEDURE TRANSPOSE used (Total process time):
       real time           0.01 seconds
       cpu time            0.01 seconds
       
 
 201              proc print noobs;
 202                 run;
 
 NOTE: There were 5 observations read from the data set WORK.DATA6.
 NOTE: PROCEDURE PRINT used (Total process time):
       real time           0.01 seconds
       cpu time            0.02 seconds
       
 
 203        
 204        proc logistic data = mathex;
 205             title3 'Multinomial logit model with proc logistic';
 206             model outcome (ref='Pass') = hsmiss / link = glogit;
 207             contrast 'HS Missing method 1' hsmiss 1;
 208             HS_MissingMethod2: test hsmiss_Fail = hsmiss_Gone = 0;
 209        
 
 NOTE: PROC LOGISTIC is fitting the generalized logit model. The logits modeled contrast each response category against the 
       reference category (outcome='Pass').
 NOTE: Convergence criterion (GCONV=1E-8) satisfied.
 NOTE: There were 579 observations read from the data set WORK.MATHEX.
 NOTE: PROCEDURE LOGISTIC used (Total process time):
       real time           0.09 seconds
       cpu time            0.08 seconds
       
 210        proc iml;
 
 NOTE: IML Ready
 211             title3 'Estimate Probabilities using output from proc catmod';
 212              b01 = -1.3594;
 212      !                      b11 = 0.6663;
 213              b02 = -0.8306;
 213      !                      b12 = 1.2360;
 214              hsmiss = 0;
 215              L1 = b01 + b11*hsmiss;
 216              L2 = b02 + b12*hsmiss;
 217              denom = 1 + exp(L1) + exp(L2);
 218              Fail = exp(L1)/denom;
 218      !                             Gone = exp(L2)/denom;
 218      !                                                   Pass = 1/denom;
 219              print "No  Missing HS Data:" Fail Gone Pass;
 220              hsmiss = 1;
 221              L1 = b01 + b11*hsmiss;
 222              L2 = b02 + b12*hsmiss;
 223              denom = 1 + exp(L1) + exp(L2);
 224              Fail = exp(L1)/denom;
 224      !                             Gone = exp(L2)/denom;
 224      !                                                   Pass = 1/denom;
 225              print "Yes Missing HS Data:" Fail Gone Pass;
 226        
 NOTE: Exiting IML.
 NOTE: PROCEDURE IML used (Total process time):
       real time           0.03 seconds
       cpu time            0.02 seconds
       
 227        proc freq  data = mathex;
 
 228             title3 'Hsmiss by outcome again for comparison';
 229             tables hsmiss * outcome / nocol nopercent;
 230        
 231        /* Now seek a good predictive model */
 232        
 
 NOTE: There were 579 observations read from the data set WORK.MATHEX.
 NOTE: PROCEDURE FREQ used (Total process time):
       real time           0.03 seconds
       cpu time            0.02 seconds
       
 233        proc logistic data = mathex;
 
 234             title3 'HS variables';
 235             model outcome (ref='Pass') = hsgpa hscalc hsengl / link = glogit;
 236        
 237        /* Drop HS English */
 238        
 
 NOTE: PROC LOGISTIC is fitting the generalized logit model. The logits modeled contrast each response category against the 
       reference category (outcome='Pass').
 NOTE: Convergence criterion (GCONV=1E-8) satisfied.
 NOTE: There were 579 observations read from the data set WORK.MATHEX.
 NOTE: PROCEDURE LOGISTIC used (Total process time):
       real time           0.10 seconds
       cpu time            0.10 seconds
       
 239        proc logistic data = mathex;
 
 240             title3 'HS gpa and calc + course2 ';
 241             class course2 / param=ref; /* Last category is reference by default. */
 242             model outcome (ref='Pass') = hsgpa hscalc course2 / link = glogit;
 243        
 244        /* Forget course2 */
 245        
 
 NOTE: PROC LOGISTIC is fitting the generalized logit model. The logits modeled contrast each response category against the 
       reference category (outcome='Pass').
 NOTE: Convergence criterion (GCONV=1E-8) satisfied.
 NOTE: There were 579 observations read from the data set WORK.MATHEX.
 NOTE: PROCEDURE LOGISTIC used (Total process time):
       real time           0.12 seconds
       cpu time            0.11 seconds
       
 246        proc logistic data = mathex;
 
 247             title3 'HS gpa and calc + diagnostic test';
 248             model outcome (ref='Pass') = hsgpa hscalc precalc calc / link = glogit;
 249        
 250        /* Drop calc subtest, keep precalc */
 251        
 
 NOTE: PROC LOGISTIC is fitting the generalized logit model. The logits modeled contrast each response category against the 
       reference category (outcome='Pass').
 NOTE: Convergence criterion (GCONV=1E-8) satisfied.
 NOTE: There were 579 observations read from the data set WORK.MATHEX.
 NOTE: PROCEDURE LOGISTIC used (Total process time):
       real time           0.10 seconds
       cpu time            0.10 seconds
       
 252        proc logistic data = mathex;
 
 253             title3 'Try gender, ethnic and mother tongue controlling for good stuff';
 254             class ethnic (param=ref ref='East Indian');
 255                  /* Specifying a reference category that's not the last value */
 256             model outcome (ref='Pass') =
 257                   hsgpa hscalc precalc ethnic gender mtongue / link = glogit;
 258             contrast 'Demographics'      ethnic  1 0 0 0 0,
 259                                          ethnic  0 1 0 0 0,
 260                                          ethnic  0 0 1 0 0,
 261                                          ethnic  0 0 0 1 0,
 262                                          ethnic  0 0 0 0 1,
 263                                          gender  1,
 264                                          mtongue 1;
 265             contrast 'Ethnic and Gender' ethnic  1 0 0 0 0,
 266                                          ethnic  0 1 0 0 0,
 267                                          ethnic  0 0 1 0 0,
 268                                          ethnic  0 0 0 1 0,
 269                                          ethnic  0 0 0 0 1,
 270                                          gender  1;
 271        
 272        /* Mother tongue is significant. Still true when we drop ethnic and gender? */
 273        
 
 NOTE: PROC LOGISTIC is fitting the generalized logit model. The logits modeled contrast each response category against the 
       reference category (outcome='Pass').
 NOTE: Convergence criterion (GCONV=1E-8) satisfied.
 NOTE: There were 579 observations read from the data set WORK.MATHEX.
 NOTE: PROCEDURE LOGISTIC used (Total process time):
       real time           0.17 seconds
       cpu time            0.16 seconds
       
 274        proc logistic data = mathex;
 
 275             title3 'hsgpa hscalc precalc mtongue';
 276             model outcome (ref='Pass') =
 277                           hsgpa hscalc precalc mtongue / link = glogit;
 278        
 
 NOTE: PROC LOGISTIC is fitting the generalized logit model. The logits modeled contrast each response category against the 
       reference category (outcome='Pass').
 NOTE: Convergence criterion (GCONV=1E-8) satisfied.
 NOTE: There were 579 observations read from the data set WORK.MATHEX.
 NOTE: PROCEDURE LOGISTIC used (Total process time):
       real time           0.10 seconds
       cpu time            0.08 seconds
       
 279        proc logistic data = mathex;
 
 280             title3 'hsgpa hscalc precalc mtongue';
 281             model outcome (ref='Pass') =
 282                           hsgpa hscalc precalc mtongue / link = glogit;
 283        
 284        /* Allowing for academic background, students whose first language is English
 285        are more likely to fail the course as opposed to passing, and less likely to
 286        disappear as opposed to passing. If this is replicated, it will be very
 287        interesting. Now explore in more detail.
 288        
 289           Recall the response categories are 1=Fail 2=Gone 3=Pass.
 290        
 291           We want to know whether failing is different from disappearing in terms
 292           of their relationship to the explanatory variables. We are getting
 293           advanced here. What is H0?
 294        
 295        Model (using b instead of beta) is
 296        
 297             ln(pi1/pi3) = b01 + b11 hsgpa + b21 hscalc + b31 precalc + b41 mtongue
 298             ln(pi2/pi3) = b02 + b12 hsgpa + b22 hscalc + b32 precalc + b42 mtongue
 299        
 300             The null hypothesis is b11=b12, b21=b22, b31=b32, b41=b42
 301        
 302        Parameter names are easy to guess.  */
 303        
 
 NOTE: PROC LOGISTIC is fitting the generalized logit model. The logits modeled contrast each response category against the 
       reference category (outcome='Pass').
 NOTE: Convergence criterion (GCONV=1E-8) satisfied.
 NOTE: There were 579 observations read from the data set WORK.MATHEX.
 NOTE: PROCEDURE LOGISTIC used (Total process time):
       real time           0.11 seconds
       cpu time            0.09 seconds
       
 304        proc logistic data = mathex;
 
 305             title3 'Different coefficients for Gone and Fail?';
 306             model outcome (ref='Pass') = hsgpa hscalc precalc mtongue / link = glogit;
 307             DiffOverall: test hsgpa_Fail = hsgpa_Gone, hscalc_Fail = hscalc_Gone,
 308                               precalc_Fail = precalc_Gone, mtongue_Fail = mtongue_Gone;
 309             Diff_hsgpa:   test hsgpa_Fail = hsgpa_Gone;
 310             Diff_hscalc:  test hscalc_Fail = hscalc_Gone;
 311             Diff_precalc: test precalc_Fail = precalc_Gone;
 312             Diff_mtongue: test mtongue_Fail = mtongue_Gone;
 313        run;
 
 NOTE: PROC LOGISTIC is fitting the generalized logit model. The logits modeled contrast each response category against the 
       reference category (outcome='Pass').
 NOTE: Convergence criterion (GCONV=1E-8) satisfied.
 NOTE: There were 579 observations read from the data set WORK.MATHEX.
 NOTE: PROCEDURE LOGISTIC used (Total process time):
       real time           0.12 seconds
       cpu time            0.09 seconds
       
 
 314        
 315        /************************** Replication ***********************
 316        For interpretation, want to replicate 8 findings:
 317        Gone vs. Pass and Fail vs. Pass for each explanatory variable.
 318        ***************************************************************/
 319        
 320        %include '/folders/myfolders/441s16/Lecture/readreplic.sas';
 NOTE: Format YNFMT is already on the library WORK.FORMATS.
 NOTE: Format YNFMT has been output.
 NOTE: Format CRSFMT is already on the library WORK.FORMATS.
 NOTE: Format CRSFMT has been output.
 NOTE: Format NFMT is already on the library WORK.FORMATS.
 NOTE: Format NFMT has been output.
 
 NOTE: PROCEDURE FORMAT used (Total process time):
       real time           0.00 seconds
       cpu time            0.01 seconds
       
 
 433        if (0<=mark<=49) then outcome = 'Fail';
 434           else if (50<=mark<=100) then outcome = 'Pass';
 435           else outcome = 'Gone';
 436        
 
 NOTE: The infile '/folders/myfolders/replicmath2.data.txt' is:
       Filename=/folders/myfolders/replicmath2.data.txt,
       Owner Name=root,Group Name=vboxsf,
       Access Permission=-rwxrwx---,
       Last Modified=30Jan2016:16:16:16,
       File Size (bytes)=38214
 
 NOTE: 579 records were read from the infile '/folders/myfolders/replicmath2.data.txt'.
       The minimum record length was 64.
       The maximum record length was 64.
 NOTE: Missing values were generated as a result of performing an operation on missing values.
       Each place is given by: (Number of times) at (Line):(Column).
       81 at 348:24   81 at 384:18   
 NOTE: The data set WORK.REPLIC has 579 observations and 35 variables.
 NOTE: DATA statement used (Total process time):
       real time           0.00 seconds
       cpu time            0.01 seconds
       
 437        proc logistic data = replic; /* That's the default anyway. */
 
 438             title2 'Replicate hsgpa hscalc precalc calc mtongue 0.05/8 = .00625';
 439             model outcome (ref='Pass') = hsgpa hscalc precalc mtongue / link = glogit;
 440             Diff_mtongue: test mtongue_Fail = mtongue_Gone;
 441        
 442        
 443        
 444        /* Final conclusions:
 445        
 446             Students with higher High School GPA were less likely to fail as
 447             opposed to passing and less likely to disappear as opposed to passing.
 448        
 449             Students with higher High School Calculus marks were less likely
 450             to disappear as opposed to passing.
 451        
 452             Students with higher scores on the pre-calculus portion of the
 453             diagnostic test were less likely to disappear as opposed to passing.
 454        
 455             There was no convincing evidence of a connection between Mother
 456             Tongue (English vs. Other) and outcome.
 457        
 458         */
 459        
 460        
 461        
 462        
 463        
 464        
 465        
 466        
 467        OPTIONS NONOTES NOSTIMER NOSOURCE NOSYNTAXCHECK;
 479