X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=src%2Ft-test.q;h=c76bd2d66937bc2801b74b5c5cc525fb240e40a8;hb=c87013b64d1731dac5da8f738ae8d1b4c1030a90;hp=3f80214494bde1013c85f05713accbe7fe6df559;hpb=ccf2f45c091ce1555b4e2a36186c501675c18a59;p=pspp-builds.git diff --git a/src/t-test.q b/src/t-test.q index 3f802144..c76bd2d6 100644 --- a/src/t-test.q +++ b/src/t-test.q @@ -36,7 +36,6 @@ #include "value-labels.h" #include "var.h" #include "vfm.h" -#include "pool.h" #include "hash.h" #include "stats.h" #include "t-test.h" @@ -56,27 +55,51 @@ /* (declarations) */ /* (functions) */ -static struct cmd_t_test cmd; +static struct cmd_t_test cmd; -static struct pool *t_test_pool ; +/* Function to use for testing for missing values */ +static is_missing_func value_is_missing; /* Variable for the GROUPS subcommand, if given. */ -static struct variable *groups; +static struct variable *indep_var; /* GROUPS: Number of values specified by the user; the values specified if any. */ -static int n_groups_values; +static int n_group_values; static union value groups_values[2]; +static enum comparison criteria[2]; + /* PAIRS: Number of pairs to be compared ; each pair. */ static int n_pairs = 0 ; struct pair { +#if 1 /* The variables comprising the pair */ struct variable *v[2]; +#endif + + /* The number of valid variable pairs */ + double n; + + /* The sum of the members */ + double sum[2]; + + /* sum of squares of the members */ + double ssq[2]; + + /* Std deviation of the members */ + double std_dev[2]; + + + /* Sample Std deviation of the members */ + double s_std_dev[2]; + + /* The means of the members */ + double mean[2]; /* The correlation coefficient between the variables */ double correlation; @@ -84,6 +107,9 @@ struct pair /* The sum of the differences */ double sum_of_diffs; + /* The sum of the products */ + double sum_of_prod; + /* The mean of the differences */ double mean_diff; @@ -93,12 +119,11 @@ struct pair /* The std deviation of the differences */ double std_dev_diff; }; -static struct pair *pairs=0; +static struct pair *pairs=0; static int parse_value (union value * v, int type) ; - /* Structures and Functions for the Statistics Summary Box */ struct ssbox; typedef void populate_ssbox_func(struct ssbox *ssb, @@ -262,25 +287,32 @@ cmd_t_test(void) } } + /* If /MISSING=INCLUDE is set, then user missing values are ignored */ + if (cmd.incl == TTS_INCLUDE ) + value_is_missing = is_system_missing; + else + value_is_missing = is_missing; - procedure(common_precalc,common_calc,common_postcalc, NULL); + procedure_with_splits (common_precalc, common_calc, common_postcalc, NULL); switch(mode) { case T_1_SAMPLE: - procedure(one_sample_precalc,one_sample_calc,one_sample_postcalc, NULL); + procedure_with_splits (one_sample_precalc, one_sample_calc, + one_sample_postcalc, NULL); break; case T_PAIRED: - procedure(paired_precalc,paired_calc,paired_postcalc, NULL); + procedure_with_splits (paired_precalc, paired_calc, paired_postcalc, + NULL); break; case T_IND_SAMPLES: - procedure(group_precalc,group_calc,group_postcalc, NULL); - levene(groups, cmd.n_variables, cmd.v_variables); + procedure_with_splits(group_precalc,group_calc,group_postcalc, NULL); + levene(indep_var, cmd.n_variables, cmd.v_variables, + (cmd.miss == TTS_LISTWISE)?LEV_LISTWISE:LEV_ANALYSIS , + value_is_missing); break; } - t_test_pool = pool_create (); - ssbox_create(&stat_summary_box,&cmd,mode); ssbox_populate(&stat_summary_box,&cmd); ssbox_finalize(&stat_summary_box); @@ -292,16 +324,10 @@ cmd_t_test(void) trbox_populate(&test_results_box,&cmd); trbox_finalize(&test_results_box); - pool_destroy (t_test_pool); - - t_test_pool=0; - - n_pairs=0; free(pairs); pairs=0; - if ( mode == T_IND_SAMPLES) { int i; @@ -318,6 +344,7 @@ cmd_t_test(void) static int tts_custom_groups (struct cmd_t_test *cmd UNUSED) { + lex_match('='); if (token != T_ALL && @@ -328,27 +355,28 @@ tts_custom_groups (struct cmd_t_test *cmd UNUSED) return 0; } - groups = parse_variable (); - if (!groups) + indep_var = parse_variable (); + if (!indep_var) { lex_error ("expecting variable name in GROUPS subcommand"); return 0; } - if (groups->type == T_STRING && groups->width > MAX_SHORT_STRING) + if (indep_var->type == T_STRING && indep_var->width > MAX_SHORT_STRING) { msg (SE, _("Long string variable %s is not valid here."), - groups->name); + indep_var->name); return 0; } if (!lex_match ('(')) { - if (groups->type == NUMERIC) + if (indep_var->type == NUMERIC) { - n_groups_values = 2; groups_values[0].f = 1; groups_values[1].f = 2; + criteria[0] = criteria[1] = CMP_EQ; + n_group_values = 2; return 1; } else @@ -359,27 +387,31 @@ tts_custom_groups (struct cmd_t_test *cmd UNUSED) } } - if (!parse_value (&groups_values[0],groups->type)) - return 0; - n_groups_values = 1; + if (!parse_value (&groups_values[0],indep_var->type)) + return 0; lex_match (','); if (lex_match (')')) - return 1; + { + criteria[0] = CMP_LE; + criteria[1] = CMP_GT; + groups_values[1] = groups_values[0]; + n_group_values = 1; + return 1; + } - if (!parse_value (&groups_values[1],groups->type)) + if (!parse_value (&groups_values[1],indep_var->type)) return 0; - n_groups_values = 2; - + + n_group_values = 2; if (!lex_force_match (')')) return 0; + criteria[0] = criteria[1] = CMP_EQ; return 1; } - - static int tts_custom_pairs (struct cmd_t_test *cmd UNUSED) { @@ -644,7 +676,7 @@ ssbox_independent_samples_init(struct ssbox *this, ssbox_base_init(this, hsize,vsize); tab_title (this->t, 0, _("Group Statistics")); tab_vline(this->t,0,1,0,vsize); - tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, groups->name); + tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, indep_var->name); tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("N")); tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("Mean")); tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation")); @@ -659,18 +691,26 @@ ssbox_independent_samples_populate(struct ssbox *ssb, { int i; + char *val_lab0=0; char *val_lab1=0; - char *val_lab2=0; - if ( groups->type == NUMERIC ) + char prefix[2][3]={"",""}; + + if ( indep_var->type == NUMERIC ) { - val_lab1 = val_labs_find( groups->val_labs,groups_values[0]); - val_lab2 = val_labs_find( groups->val_labs,groups_values[1]); + val_lab0 = val_labs_find( indep_var->val_labs,groups_values[0]); + val_lab1 = val_labs_find( indep_var->val_labs,groups_values[1]); } else { - val_lab1 = groups_values[0].s; - val_lab2 = groups_values[1].s; + val_lab0 = groups_values[0].s; + val_lab1 = groups_values[1].s; + } + + if (n_group_values == 1) + { + strcpy(prefix[0],"< "); + strcpy(prefix[1],">="); } assert(ssb->t); @@ -681,16 +721,20 @@ ssbox_independent_samples_populate(struct ssbox *ssb, tab_text (ssb->t, 0, i*2+1, TAB_LEFT, cmd->v_variables[i]->name); - if (val_lab1) - tab_text (ssb->t, 1, i*2+1, TAB_LEFT, val_lab1); + if (val_lab0) + tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF, + "%s%s", prefix[0], val_lab0); else - tab_float(ssb->t, 1 ,i*2+1, TAB_LEFT, groups_values[0].f, 2,0); + tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF, + "%s%g", prefix[0], groups_values[0].f); - if (val_lab2) - tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT, val_lab2); + if (val_lab1) + tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF, + "%s%s", prefix[1], val_lab1); else - tab_float(ssb->t, 1 ,i*2+1+1, TAB_LEFT, groups_values[1].f,2,0); + tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF, + "%s%g", prefix[1], groups_values[1].f); /* Fill in the group statistics */ for ( g=0; g < 2 ; ++g ) @@ -755,15 +799,13 @@ ssbox_paired_populate(struct ssbox *ssb,struct cmd_t_test *cmd UNUSED) tab_text (ssb->t, 1, i*2+j+1, TAB_LEFT, pairs[i].v[j]->name); /* Values */ - tab_float (ssb->t,2, i*2+j+1, TAB_RIGHT, gs->mean, 8, 2); - tab_float (ssb->t,3, i*2+j+1, TAB_RIGHT, gs->n, 2, 0); - tab_float (ssb->t,4, i*2+j+1, TAB_RIGHT, gs->std_dev, 8, 3); - tab_float (ssb->t,5, i*2+j+1, TAB_RIGHT, gs->se_mean, 8, 3); + tab_float (ssb->t,2, i*2+j+1, TAB_RIGHT, pairs[i].mean[j], 8, 2); + tab_float (ssb->t,3, i*2+j+1, TAB_RIGHT, pairs[i].n, 2, 0); + tab_float (ssb->t,4, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j], 8, 3); + tab_float (ssb->t,5, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j]/sqrt(pairs[i].n), 8, 3); } - } - } /* Populate the one sample ssbox */ @@ -1097,13 +1139,7 @@ trbox_paired_populate(struct trbox *trb, double bound; double se_mean; - struct variable *v0 = pairs[i].v[0]; - struct variable *v1 = pairs[i].v[1]; - - struct group_statistics *gs0 = &v0->p.t_t.ugs; - struct group_statistics *gs1 = &v1->p.t_t.ugs; - - double n = gs0->n; + double n = pairs[i].n; double t; double df = n - 1; @@ -1137,12 +1173,13 @@ trbox_paired_populate(struct trbox *trb, tab_float(trb->t, 6, i+3, TAB_RIGHT, pairs[i].mean_diff + t * se_mean , 8, 4); - t = ( gs0->mean - gs1->mean) - / sqrt ( - ( sqr(gs0->s_std_dev) + sqr(gs1->s_std_dev) - - 2 * pairs[i].correlation * gs0->s_std_dev * gs1->s_std_dev ) - / (n-1) ) - ; + t = (pairs[i].mean[0] - pairs[i].mean[1]) + / sqrt ( + ( sqr (pairs[i].s_std_dev[0]) + sqr (pairs[i].s_std_dev[1]) - + 2 * pairs[i].correlation * + pairs[i].s_std_dev[0] * pairs[i].s_std_dev[1] ) + / (n - 1) + ); tab_float(trb->t, 7, i+3, TAB_RIGHT, t , 8,3 ); @@ -1151,7 +1188,6 @@ trbox_paired_populate(struct trbox *trb, which=1; cdft(&which, &p, &q, &t, &df, &status, &bound); - if ( 0 != status ) { msg( SE, _("Error calculating T statistic (cdft returned %d)."),status); @@ -1318,7 +1354,7 @@ pscbox(void) int status; double bound; - double df = pairs[i].v[0]->p.t_t.ugs.n -2; + double df = pairs[i].n -2; double correlation_t = pairs[i].correlation * sqrt(df) / @@ -1334,20 +1370,16 @@ pscbox(void) /* row data */ + tab_float(table, 2, i+1, TAB_RIGHT, pairs[i].n, 4, 0); tab_float(table, 3, i+1, TAB_RIGHT, pairs[i].correlation, 8, 3); - tab_float(table, 2, i+1, TAB_RIGHT, pairs[i].v[0]->p.t_t.ugs.n , 4, 0); - cdft(&which, &p, &q, &correlation_t, &df, &status, &bound); - if ( 0 != status ) { msg( SE, _("Error calculating T statistic (cdft returned %d)."),status); } - tab_float(table, 4, i+1, TAB_RIGHT, 2.0*(correlation_t>0?q:p), 8, 3); - } tab_submit(table); @@ -1365,6 +1397,33 @@ common_calc (struct ccase *c, void *aux UNUSED) double weight = dict_get_case_weight(default_dict,c); + + /* Skip the entire case if /MISSING=LISTWISE is set */ + if ( cmd.miss == TTS_LISTWISE ) + { + for(i=0; i< cmd.n_variables ; ++i) + { + struct variable *v = cmd.v_variables[i]; + union value *val = &c->data[v->fv]; + + if (value_is_missing(val,v) ) + { + return 0; + } + } + } + + /* Listwise has to be implicit if the independent variable is missing ?? */ + if ( cmd.sbc_groups ) + { + union value *gv = &c->data[indep_var->fv]; + if ( value_is_missing(gv,indep_var) ) + { + return 0; + } + } + + for(i=0; i< cmd.n_variables ; ++i) { struct group_statistics *gs; @@ -1373,7 +1432,7 @@ common_calc (struct ccase *c, void *aux UNUSED) gs= &cmd.v_variables[i]->p.t_t.ugs; - if (val->f != SYSMIS) + if (! value_is_missing(val,v) ) { gs->n+=weight; gs->sum+=weight * val->f; @@ -1435,6 +1494,21 @@ one_sample_calc (struct ccase *c, void *aux UNUSED) double weight = dict_get_case_weight(default_dict,c); + /* Skip the entire case if /MISSING=LISTWISE is set */ + if ( cmd.miss == TTS_LISTWISE ) + { + for(i=0; i< cmd.n_variables ; ++i) + { + struct variable *v = cmd.v_variables[i]; + union value *val = &c->data[v->fv]; + + if (value_is_missing(val,v) ) + { + return 0; + } + } + } + for(i=0; i< cmd.n_variables ; ++i) { struct group_statistics *gs; @@ -1443,7 +1517,7 @@ one_sample_calc (struct ccase *c, void *aux UNUSED) gs= &cmd.v_variables[i]->p.t_t.ugs; - if (val->f != SYSMIS) + if ( ! value_is_missing(val,v)) gs->sum_diff += weight * (val->f - cmd.n_testval); } @@ -1509,9 +1583,13 @@ paired_precalc (void *aux UNUSED) for(i=0; i < n_pairs ; ++i ) { - pairs[i].correlation=0; - pairs[i].sum_of_diffs=0; - pairs[i].ssq_diffs=0; + pairs[i].n = 0; + pairs[i].sum[0] = 0; pairs[i].sum[1] = 0; + pairs[i].ssq[0] = 0; pairs[i].ssq[1] = 0; + pairs[i].sum_of_prod = 0; + pairs[i].correlation = 0; + pairs[i].sum_of_diffs = 0; + pairs[i].ssq_diffs = 0; } } @@ -1522,6 +1600,28 @@ paired_calc (struct ccase *c, void *aux UNUSED) { int i; + double weight = dict_get_case_weight(default_dict,c); + + /* Skip the entire case if /MISSING=LISTWISE is set , + AND one member of a pair is missing */ + if ( cmd.miss == TTS_LISTWISE ) + { + for(i=0; i < n_pairs ; ++i ) + { + struct variable *v0 = pairs[i].v[0]; + struct variable *v1 = pairs[i].v[1]; + + union value *val0 = &c->data[v0->fv]; + union value *val1 = &c->data[v1->fv]; + + if ( value_is_missing(val0,v0) || + value_is_missing(val1,v1) ) + { + return 0; + } + } + } + for(i=0; i < n_pairs ; ++i ) { struct variable *v0 = pairs[i].v[0]; @@ -1530,13 +1630,28 @@ paired_calc (struct ccase *c, void *aux UNUSED) union value *val0 = &c->data[v0->fv]; union value *val1 = &c->data[v1->fv]; - pairs[i].correlation += ( val0->f - pairs[i].v[0]->p.t_t.ugs.mean ) - * - ( val1->f - pairs[i].v[1]->p.t_t.ugs.mean ); + if ( ( !value_is_missing(val0,v0) && !value_is_missing(val1,v1) ) ) + { + pairs[i].n += weight; + pairs[i].sum[0] += weight * val0->f; + pairs[i].sum[1] += weight * val1->f; + + pairs[i].ssq[0] += weight * sqr(val0->f); + pairs[i].ssq[1] += weight * sqr(val1->f); + +#if 0 + pairs[i].correlation += weight * + ( val0->f - pairs[i].v[0]->p.t_t.ugs.mean ) + * + ( val1->f - pairs[i].v[1]->p.t_t.ugs.mean ); +#endif - pairs[i].sum_of_diffs += val0->f - val1->f ; - pairs[i].ssq_diffs += sqr(val0->f - val1->f); + pairs[i].sum_of_prod += weight * val0->f * val1->f ; + + pairs[i].sum_of_diffs += weight * ( val0->f - val1->f ) ; + pairs[i].ssq_diffs += weight * sqr(val0->f - val1->f); + } } return 0; @@ -1549,11 +1664,33 @@ paired_postcalc (void *aux UNUSED) for(i=0; i < n_pairs ; ++i ) { - const double n = pairs[i].v[0]->p.t_t.ugs.n ; + int j; + const double n = pairs[i].n; + + for (j=0; j < 2 ; ++j) + { + pairs[i].mean[j] = pairs[i].sum[j] / n ; + pairs[i].s_std_dev[j] = sqrt((pairs[i].ssq[j] / n - + sqr(pairs[i].mean[j])) + ); + + pairs[i].std_dev[j] = sqrt(n/(n-1)*(pairs[i].ssq[j] / n - + sqr(pairs[i].mean[j])) + ); + } + + pairs[i].correlation = pairs[i].sum_of_prod / pairs[i].n - + pairs[i].mean[0] * pairs[i].mean[1] ; + /* correlation now actually contains the covariance */ + pairs[i].correlation /= pairs[i].std_dev[0] * pairs[i].std_dev[1]; + pairs[i].correlation *= pairs[i].n / ( pairs[i].n - 1 ); + +#if 0 pairs[i].correlation /= pairs[i].v[0]->p.t_t.ugs.std_dev * pairs[i].v[1]->p.t_t.ugs.std_dev ; - pairs[i].correlation /= pairs[i].v[0]->p.t_t.ugs.n -1; + pairs[i].correlation /= n - 1; +#endif pairs[i].mean_diff = pairs[i].sum_of_diffs / n ; @@ -1567,16 +1704,40 @@ paired_postcalc (void *aux UNUSED) } } +/* Return the group # corresponding to the + independent variable with the value val +*/ static int -get_group(const union value *val, struct variable *var) +get_group(const union value *val, struct variable *indep) { - if ( 0 == compare_values(val,&groups_values[0],var->width) ) - return 0; - else if (0 == compare_values(val,&groups_values[1],var->width) ) - return 1; + int i; - /* Never reached */ - assert(0); + for (i = 0; i < 2 ; ++i ) + { + const int cmp = compare_values(val,&groups_values[i],indep->width) ; + switch ( criteria[i]) + { + case CMP_EQ: + if ( 0 == cmp ) return i; + break; + case CMP_LT: + if ( 0 > cmp ) return i; + break; + case CMP_LE: + if ( cmp <= 0 ) return i; + break; + case CMP_GT: + if ( cmp > 0 ) return i; + break; + case CMP_GE: + if ( cmp >= 0 ) return i; + break; + default: + assert(0); + }; + } + + /* No groups matched */ return -1; } @@ -1597,10 +1758,15 @@ group_precalc (void *aux UNUSED) for (j=0 ; j < 2 ; ++j) { - ttpr->gs[j].sum=0; - ttpr->gs[j].n=0; - ttpr->gs[j].ssq=0; - ttpr->gs[j].id = groups_values[j]; + ttpr->gs[j].sum = 0; + ttpr->gs[j].n = 0; + ttpr->gs[j].ssq = 0; + + if ( n_group_values == 2 ) + ttpr->gs[j].id = groups_values[j]; + else + ttpr->gs[j].id = groups_values[0]; + ttpr->gs[j].criterion = criteria[j]; } } @@ -1610,23 +1776,54 @@ static int group_calc (struct ccase *c, void *aux UNUSED) { int i; - union value *gv = &c->data[groups->fv]; + int g; + union value *gv = &c->data[indep_var->fv]; double weight = dict_get_case_weight(default_dict,c); - gv = &c->data[groups->fv]; + if ( value_is_missing(gv,indep_var) ) + { + return 0; + } + + if ( cmd.miss == TTS_LISTWISE ) + { + for(i=0; i< cmd.n_variables ; ++i) + { + struct variable *v = cmd.v_variables[i]; + union value *val = &c->data[v->fv]; + + if (value_is_missing(val,v) ) + { + return 0; + } + } + } + + + gv = &c->data[indep_var->fv]; + + g = get_group(gv,indep_var); + + /* If the independent variable doesn't match either of the values + for this case then move on to the next case */ + if (g == -1 ) + return 0; for(i=0; i< cmd.n_variables ; ++i) { - int g = get_group(gv,groups); + struct variable *var = cmd.v_variables[i]; - struct group_statistics *gs = &cmd.v_variables[i]->p.t_t.gs[g]; + struct group_statistics *gs = &var->p.t_t.gs[g]; - union value *val=&c->data[cmd.v_variables[i]->fv]; + union value *val=&c->data[var->fv]; - gs->n+=weight; - gs->sum+=weight * val->f; - gs->ssq+=weight * sqr(val->f); + if ( !value_is_missing(val,var) ) + { + gs->n+=weight; + gs->sum+=weight * val->f; + gs->ssq+=weight * sqr(val->f); + } } return 0;