X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=src%2Foneway.q;h=075ce0a1f00b04513481e84530ff22c71ae73244;hb=1fc3af93c0ba6cbaf7ef09edc979096b6f16dd6f;hp=17476cc786a827af69222dfd08f794277a356345;hpb=205eaea8e2d95e20baa2c00a495b0ac4f9646372;p=pspp diff --git a/src/oneway.q b/src/oneway.q index 17476cc786..075ce0a1f0 100644 --- a/src/oneway.q +++ b/src/oneway.q @@ -39,7 +39,7 @@ #include "vfm.h" #include "hash.h" #include "casefile.h" -#include "oneway.h" +#include "group_proc.h" #include "group.h" #include "levene.h" @@ -95,6 +95,10 @@ static void show_contrast_coeffs(void); static void show_contrast_tests(void); +enum stat_table_t {STAT_DESC = 1, STAT_HOMO = 2}; + +static enum stat_table_t stat_tables ; + int cmd_oneway(void) @@ -110,6 +114,25 @@ cmd_oneway(void) else value_is_missing = is_missing; + /* What statistics were requested */ + if ( cmd.sbc_statistics ) + { + + for (i = 0 ; i < ONEWAY_ST_count ; ++i ) + { + if ( ! cmd.a_statistics[i] ) continue; + + switch (i) { + case ONEWAY_ST_DESCRIPTIVES: + stat_tables |= STAT_DESC; + break; + case ONEWAY_ST_HOMOGENEITY: + stat_tables |= STAT_HOMO; + break; + } + } + } + multipass_procedure_with_splits (calculate, &cmd); /* Check the sanity of the given contrast values */ @@ -133,26 +156,11 @@ cmd_oneway(void) msg(SW,_("Coefficients for contrast %d do not total zero"),i + 1); } + if ( stat_tables & STAT_DESC ) + show_descriptives(); - - /* Show the statistics tables */ - if ( cmd.sbc_statistics ) - { - for (i = 0 ; i < ONEWAY_ST_count ; ++i ) - { - if ( ! cmd.a_statistics[i] ) continue; - - switch (i) { - case ONEWAY_ST_DESCRIPTIVES: - show_descriptives(); - break; - case ONEWAY_ST_HOMOGENEITY: - show_homogeneity(); - break; - } - } - } - + if ( stat_tables & STAT_HOMO ) + show_homogeneity(); show_anova_table(); @@ -166,7 +174,7 @@ cmd_oneway(void) /* Clean up */ for (i = 0 ; i < n_vars ; ++i ) { - struct hsh_table *group_hash = vars[i]->p.ww.group_hash; + struct hsh_table *group_hash = vars[i]->p.grp_data.group_hash; hsh_destroy(group_hash); } @@ -254,8 +262,8 @@ show_anova_table(void) for ( i=0 ; i < n_vars ; ++i ) { - struct group_statistics *totals = &vars[i]->p.ww.ugs; - struct hsh_table *group_hash = vars[i]->p.ww.group_hash; + struct group_statistics *totals = &vars[i]->p.grp_data.ugs; + struct hsh_table *group_hash = vars[i]->p.grp_data.group_hash; struct hsh_iterator g; struct group_statistics *gs; double ssa=0; @@ -283,11 +291,11 @@ show_anova_table(void) { const double sst = totals->ssq - ( totals->sum * totals->sum) / totals->n ; - const double df1 = vars[i]->p.ww.n_groups - 1; - const double df2 = totals->n - vars[i]->p.ww.n_groups ; + const double df1 = vars[i]->p.grp_data.n_groups - 1; + const double df2 = totals->n - vars[i]->p.grp_data.n_groups ; const double msa = ssa / df1; - vars[i]->p.ww.mse = (sst - ssa) / df2; + vars[i]->p.grp_data.mse = (sst - ssa) / df2; /* Sums of Squares */ @@ -303,11 +311,11 @@ show_anova_table(void) /* Mean Squares */ tab_float (t, 4, i * 3 + 1, TAB_RIGHT, msa, 8, 3); - tab_float (t, 4, i * 3 + 2, TAB_RIGHT, vars[i]->p.ww.mse, 8, 3); + tab_float (t, 4, i * 3 + 2, TAB_RIGHT, vars[i]->p.grp_data.mse, 8, 3); { - const double F = msa/vars[i]->p.ww.mse ; + const double F = msa/vars[i]->p.grp_data.mse ; /* The F value */ tab_float (t, 5, i * 3 + 1, 0, F, 8, 3); @@ -345,7 +353,7 @@ show_descriptives(void) for ( v = 0 ; v < n_vars ; ++v ) - n_rows += vars[v]->p.ww.n_groups + 1; + n_rows += vars[v]->p.grp_data.n_groups + 1; t = tab_create (n_cols,n_rows,0); tab_headers (t, 2, 0, 2, 0); @@ -387,17 +395,17 @@ show_descriptives(void) for ( v=0 ; v < n_vars ; ++v ) { double T; - double stderr; + double std_error; struct hsh_iterator g; struct group_statistics *gs; - struct group_statistics *totals = &vars[v]->p.ww.ugs; + struct group_statistics *totals = &vars[v]->p.grp_data.ugs; int count = 0 ; char *s = (vars[v]->label) ? vars[v]->label : vars[v]->name; - struct hsh_table *group_hash = vars[v]->p.ww.group_hash; + struct hsh_table *group_hash = vars[v]->p.grp_data.group_hash; tab_text (t, 0, row, TAB_LEFT | TAT_TITLE, s); @@ -430,19 +438,19 @@ show_descriptives(void) tab_float (t, 4, row + count, 0, gs->std_dev,8,2); - stderr = gs->std_dev/sqrt(gs->n) ; + std_error = gs->std_dev/sqrt(gs->n) ; tab_float (t, 5, row + count, 0, - stderr, 8,2); + std_error, 8,2); /* Now the confidence interval */ T = gsl_cdf_tdist_Qinv(q,gs->n - 1); tab_float(t, 6, row + count, 0, - gs->mean - T * stderr, 8, 2); + gs->mean - T * std_error, 8, 2); tab_float(t, 7, row + count, 0, - gs->mean + T * stderr, 8, 2); + gs->mean + T * std_error, 8, 2); /* Min and Max */ @@ -461,26 +469,26 @@ show_descriptives(void) tab_float (t, 4, row + count, 0, totals->std_dev,8,2); - stderr = totals->std_dev/sqrt(totals->n) ; + std_error = totals->std_dev/sqrt(totals->n) ; - tab_float (t, 5, row + count, 0, stderr, 8,2); + tab_float (t, 5, row + count, 0, std_error, 8,2); /* Now the confidence interval */ T = gsl_cdf_tdist_Qinv(q,totals->n - 1); tab_float(t, 6, row + count, 0, - totals->mean - T * stderr, 8, 2); + totals->mean - T * std_error, 8, 2); tab_float(t, 7, row + count, 0, - totals->mean + T * stderr, 8, 2); + totals->mean + T * std_error, 8, 2); /* Min and Max */ tab_float(t, 8, row + count, 0, totals->minimum, 8, 2); tab_float(t, 9, row + count, 0, totals->maximum, 8, 2); - row += vars[v]->p.ww.n_groups + 1; + row += vars[v]->p.grp_data.n_groups + 1; } @@ -526,9 +534,23 @@ show_homogeneity(void) for ( v=0 ; v < n_vars ; ++v ) { - char *s = (vars[v]->label) ? vars[v]->label : vars[v]->name; + double F; + const struct variable *var = vars[v]; + const char *s = (var->label) ? var->label : var->name; + const struct group_statistics *totals = &var->p.grp_data.ugs; + + const double df1 = var->p.grp_data.n_groups - 1; + const double df2 = totals->n - var->p.grp_data.n_groups ; tab_text (t, 0, v + 1, TAB_LEFT | TAT_TITLE, s); + + F = var->p.grp_data.levene; + tab_float (t, 1, v + 1, TAB_RIGHT, F, 8,3); + tab_float (t, 2, v + 1, TAB_RIGHT, df1 ,8,0); + tab_float (t, 3, v + 1, TAB_RIGHT, df2 ,8,0); + + /* Now the significance */ + tab_float (t, 4, v + 1, TAB_RIGHT,gsl_cdf_fdist_Q(F,df1,df2), 8, 3); } tab_submit (t); @@ -680,14 +702,30 @@ show_contrast_tests(void) int ci; double contrast_value = 0.0; double coef_msq = 0.0; - struct oneway_proc *ww = &vars[v]->p.ww ; - struct hsh_table *group_hash = ww->group_hash; + struct group_proc *grp_data = &vars[v]->p.grp_data ; + struct hsh_table *group_hash = grp_data->group_hash; struct hsh_iterator g; struct group_statistics *gs; double T; - double stderr_contrast ; + double std_error_contrast ; double df; + double sec_vneq=0.0; + + + /* Note: The calculation of the degrees of freedom in the variances + not equal case is painfull!! + The following formula may help to understand it: + \frac{\left(\sum_{i=1}^k{c_i^2\frac{s_i^2}{n_i}}\right)^2} + { + \sum_{i=1}^k\left( + \frac{\left(c_i^2\frac{s_i^2}{n_i}\right)^2} {n_i-1} + \right) + } + */ + + double df_denominator = 0.0; + double df_numerator = 0.0; if ( i == 0 ) @@ -715,12 +753,25 @@ show_contrast_tests(void) gs != 0; ++ci, gs = hsh_next(group_hash,&g)) { + const double coef = subc_list_double_at(&cmd.dl_contrast[i],ci); + const double winv = (gs->std_dev * gs->std_dev) / gs->n; contrast_value += coef * gs->mean; coef_msq += (coef * coef) / gs->n ; + + sec_vneq += (coef * coef) * (gs->std_dev * gs->std_dev ) /gs->n ; + + df_numerator += (coef * coef) * winv; + df_denominator += pow2((coef * coef) * winv) / (gs->n - 1); + } + sec_vneq = sqrt(sec_vneq); + + + df_numerator = pow2(df_numerator); + tab_float (t, 3, (v * lines_per_variable) + i + 1, TAB_RIGHT, contrast_value, 8,2); @@ -729,14 +780,14 @@ show_contrast_tests(void) TAB_RIGHT, contrast_value, 8,2); - stderr_contrast = sqrt(vars[v]->p.ww.mse * coef_msq); + std_error_contrast = sqrt(vars[v]->p.grp_data.mse * coef_msq); /* Std. Error */ tab_float (t, 4, (v * lines_per_variable) + i + 1, - TAB_RIGHT, stderr_contrast, + TAB_RIGHT, std_error_contrast, 8,3); - T = fabs(contrast_value / stderr_contrast) ; + T = fabs(contrast_value / std_error_contrast) ; /* T Statistic */ @@ -744,7 +795,7 @@ show_contrast_tests(void) TAB_RIGHT, T, 8,3); - df = ww->ugs.n - ww->n_groups; + df = grp_data->ugs.n - grp_data->n_groups; /* Degrees of Freedom */ tab_float (t, 6, (v * lines_per_variable) + i + 1, @@ -757,51 +808,54 @@ show_contrast_tests(void) TAB_RIGHT, 2 * gsl_cdf_tdist_Q(T,df), 8,3); - } - if ( v > 0 ) - tab_hline(t, TAL_1, 0, n_cols - 1, (v * lines_per_variable) + 1); - } + /* Now for the Variances NOT Equal case */ - tab_submit (t); + /* Std. Error */ + tab_float (t, 4, + (v * lines_per_variable) + i + 1 + cmd.sbc_contrast, + TAB_RIGHT, sec_vneq, + 8,3); -} + T = contrast_value / sec_vneq; + tab_float (t, 5, + (v * lines_per_variable) + i + 1 + cmd.sbc_contrast, + TAB_RIGHT, T, + 8,3); -/* ONEWAY ANOVA Calculations */ -static void postcalc ( struct cmd_oneway *cmd UNUSED ); + df = df_numerator / df_denominator; -static void precalc ( struct cmd_oneway *cmd UNUSED ); + tab_float (t, 6, + (v * lines_per_variable) + i + 1 + cmd.sbc_contrast, + TAB_RIGHT, df, + 8,3); -int compare_group_id (const struct group_statistics *a, - const struct group_statistics *b, int width); + /* The Significance */ -unsigned int hash_group_id(const struct group_statistics *v, int width); + tab_float (t, 7, (v * lines_per_variable) + i + 1 + cmd.sbc_contrast, + TAB_RIGHT, 2 * gsl_cdf_tdist_Q(T,df), + 8,3); -void free_group_id(struct group_statistics *v, void *aux UNUSED); + } + if ( v > 0 ) + tab_hline(t, TAL_1, 0, n_cols - 1, (v * lines_per_variable) + 1); + } + tab_submit (t); -int -compare_group_id (const struct group_statistics *a, - const struct group_statistics *b, int width) -{ - return compare_values(&a->id, &b->id, width); } -unsigned int -hash_group_id(const struct group_statistics *v, int width) -{ - return hash_value ( &v->id, width); -} -void -free_group_id(struct group_statistics *v, void *aux UNUSED) -{ - free(v); -} +/* ONEWAY ANOVA Calculations */ + +static void postcalc ( struct cmd_oneway *cmd UNUSED ); + +static void precalc ( struct cmd_oneway *cmd UNUSED ); + /* Pre calculations */ @@ -812,17 +866,17 @@ precalc ( struct cmd_oneway *cmd UNUSED ) for(i=0; i< n_vars ; ++i) { - struct group_statistics *totals = &vars[i]->p.ww.ugs; + struct group_statistics *totals = &vars[i]->p.grp_data.ugs; /* Create a hash for each of the dependent variables. The hash contains a group_statistics structure, and is keyed by value of the independent variable */ - vars[i]->p.ww.group_hash = + vars[i]->p.grp_data.group_hash = hsh_create(4, - (hsh_compare_func *) compare_group_id, - (hsh_hash_func *) hash_group_id, - (hsh_free_func *) free_group_id, + (hsh_compare_func *) compare_group, + (hsh_hash_func *) hash_group, + (hsh_free_func *) free_group, (void *) indep_var->width ); @@ -842,7 +896,6 @@ calculate(const struct casefile *cf, void *cmd_) struct casereader *r; struct ccase c; - struct cmd_oneway *cmd = (struct cmd_oneway *) cmd_; global_group_hash = hsh_create(4, @@ -850,7 +903,6 @@ calculate(const struct casefile *cf, void *cmd_) (hsh_hash_func *) hash_value, 0, (void *) indep_var->width ); - precalc(cmd); for(r = casefile_get_reader (cf); @@ -863,17 +915,37 @@ calculate(const struct casefile *cf, void *cmd_) dict_get_case_weight(default_dict,&c,&bad_weight_warn); const union value *indep_val = case_data (&c, indep_var->fv); + + /* Deal with missing values */ + if ( value_is_missing(indep_val,indep_var) ) + continue; + + /* Skip the entire case if /MISSING=LISTWISE is set */ + if ( cmd->miss == ONEWAY_LISTWISE ) + { + for(i = 0; i < n_vars ; ++i) + { + const struct variable *v = vars[i]; + const union value *val = case_data (&c, v->fv); + + if (value_is_missing(val,v) ) + break; + } + if ( i != n_vars ) + continue; + + } + hsh_insert ( global_group_hash, (void *) indep_val ); - for ( i = 0 ; i < n_vars ; ++i ) { const struct variable *v = vars[i]; const union value *val = case_data (&c, v->fv); - struct hsh_table *group_hash = vars[i]->p.ww.group_hash; + struct hsh_table *group_hash = vars[i]->p.grp_data.group_hash; struct group_statistics *gs; @@ -897,7 +969,7 @@ calculate(const struct casefile *cf, void *cmd_) if (! value_is_missing(val,v) ) { - struct group_statistics *totals = &vars[i]->p.ww.ugs; + struct group_statistics *totals = &vars[i]->p.grp_data.ugs; totals->n+=weight; totals->sum+=weight * val->f; @@ -920,7 +992,7 @@ calculate(const struct casefile *cf, void *cmd_) gs->maximum = val->f * weight; } - vars[i]->p.ww.n_groups = hsh_count ( group_hash ); + vars[i]->p.grp_data.n_groups = hsh_count ( group_hash ); } } @@ -928,6 +1000,12 @@ calculate(const struct casefile *cf, void *cmd_) postcalc(cmd); + + if ( stat_tables & STAT_HOMO ) + levene(cf, indep_var, n_vars, vars, + (cmd->miss == ONEWAY_LISTWISE) ? LEV_LISTWISE : LEV_ANALYSIS , + value_is_missing); + ostensible_number_of_groups = hsh_count (global_group_hash); } @@ -942,8 +1020,8 @@ postcalc ( struct cmd_oneway *cmd UNUSED ) for(i = 0; i < n_vars ; ++i) { - struct hsh_table *group_hash = vars[i]->p.ww.group_hash; - struct group_statistics *totals = &vars[i]->p.ww.ugs; + struct hsh_table *group_hash = vars[i]->p.grp_data.group_hash; + struct group_statistics *totals = &vars[i]->p.grp_data.ugs; struct hsh_iterator g; struct group_statistics *gs; @@ -977,8 +1055,5 @@ postcalc ( struct cmd_oneway *cmd UNUSED ) totals->se_mean = totals->std_dev / sqrt(totals->n); - - - } }