X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=src%2Foneway.q;h=ff6b79b66e9d322eac109ce062ac5806cf8bfef5;hb=1143173e5e7e57d9020a0b3303c980e8166b3642;hp=57fefc7f227021a954f9b644bd77f8ac857cb471;hpb=0a2cdacebb16d9a13c1b382c5ffa29e32cf858ea;p=pspp-builds.git diff --git a/src/oneway.q b/src/oneway.q index 57fefc7f..ff6b79b6 100644 --- a/src/oneway.q +++ b/src/oneway.q @@ -39,6 +39,8 @@ #include "vfm.h" #include "hash.h" #include "casefile.h" +#include "group_proc.h" +#include "group.h" #include "levene.h" /* (specification) @@ -54,17 +56,14 @@ +static int bad_weight_warn = 1; + + static struct cmd_oneway cmd; /* The independent variable */ static struct variable *indep_var; -/* A hash of the values of the independent variable */ -struct hsh_table *ind_vals; - -/* Number of factors (groups) */ -static int n_groups; - /* Number of dependent variables */ static int n_vars; @@ -72,24 +71,37 @@ static int n_vars; static struct variable **vars; +/* A hash table containing all the distinct values of the independent + variables */ +static struct hsh_table *global_group_hash ; +/* The number of distinct values of the independent variable, when all + missing values are disregarded */ +static int ostensible_number_of_groups=-1; /* Function to use for testing for missing values */ static is_missing_func value_is_missing; -static void calculate(const struct casefile *cf, void *_mode); +static void run_oneway(const struct casefile *cf, void *_mode); /* Routines to show the output tables */ static void show_anova_table(void); static void show_descriptives(void); static void show_homogeneity(void); -static void show_contrast_coeffs(void); -static void show_contrast_tests(void); + +static void show_contrast_coeffs(short *); +static void show_contrast_tests(short *); +enum stat_table_t {STAT_DESC = 1, STAT_HOMO = 2}; + +static enum stat_table_t stat_tables ; + +void output_oneway(void); + int cmd_oneway(void) @@ -105,7 +117,40 @@ cmd_oneway(void) else value_is_missing = is_missing; - multipass_procedure_with_splits (calculate, &cmd); + /* 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 (run_oneway, &cmd); + + + return CMD_SUCCESS; +} + + +void +output_oneway(void) +{ + + int i; + short *bad_contrast ; + + bad_contrast = xmalloc ( sizeof (short) * cmd.sbc_contrast ); /* Check the sanity of the given contrast values */ for (i = 0 ; i < cmd.sbc_contrast ; ++i ) @@ -113,52 +158,51 @@ cmd_oneway(void) int j; double sum = 0; - if ( subc_list_double_count(&cmd.dl_contrast[i]) != n_groups ) + bad_contrast[i] = 0; + if ( subc_list_double_count(&cmd.dl_contrast[i]) != + ostensible_number_of_groups ) { - msg(SE, _("Number of contrast coefficients must equal the number of groups")); - return CMD_FAILURE; + msg(SW, + _("Number of contrast coefficients must equal the number of groups")); + bad_contrast[i] = 1; + continue; } - for (j=0; j < n_groups ; ++j ) + for (j=0; j < ostensible_number_of_groups ; ++j ) sum += subc_list_double_at(&cmd.dl_contrast[i],j); if ( sum != 0.0 ) 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(); - if (cmd.sbc_contrast) + if (cmd.sbc_contrast ) { - show_contrast_coeffs(); - show_contrast_tests(); + show_contrast_coeffs(bad_contrast); + show_contrast_tests(bad_contrast); } - hsh_destroy(ind_vals); - return CMD_SUCCESS; -} + free(bad_contrast); + /* Clean up */ + for (i = 0 ; i < n_vars ; ++i ) + { + struct hsh_table *group_hash = vars[i]->p.grp_data.group_hash; + + hsh_destroy(group_hash); + } + + hsh_destroy(global_group_hash); + +} @@ -237,7 +281,24 @@ show_anova_table(void) for ( i=0 ; i < n_vars ; ++i ) { - char *s = (vars[i]->label) ? vars[i]->label : vars[i]->name; + 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; + + + for (gs = hsh_first (group_hash,&g); + gs != 0; + gs = hsh_next(group_hash,&g)) + { + ssa += (gs->sum * gs->sum)/gs->n; + } + + ssa -= ( totals->sum * totals->sum ) / totals->n ; + + const char *s = (vars[i]->label) ? vars[i]->label : vars[i]->name; + tab_text (t, 0, i * 3 + 1, TAB_LEFT | TAT_TITLE, s); tab_text (t, 1, i * 3 + 1, TAB_LEFT | TAT_TITLE, _("Between Groups")); @@ -246,68 +307,73 @@ show_anova_table(void) if (i > 0) tab_hline(t, TAL_1, 0, n_cols - 1 , i * 3 + 1); - } - - tab_title (t, 0, "ANOVA"); - tab_submit (t); - - -} - - -static void -calculate(const struct casefile *cf, void *cmd_) -{ - struct casereader *r; - struct ccase c; - - struct cmd_t_test *cmd = (struct cmd_t_test *) cmd_; - - - ind_vals = hsh_create(4, (hsh_compare_func *) compare_values, - (hsh_hash_func *) hash_value, - 0, (void *) indep_var->width ); - - for(r = casefile_get_reader (cf); - casereader_read (r, &c) ; - case_destroy (&c)) - { + { + const double sst = totals->ssq - ( totals->sum * totals->sum) / totals->n ; + 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.grp_data.mse = (sst - ssa) / df2; + + + /* Sums of Squares */ + tab_float (t, 2, i * 3 + 1, 0, ssa, 10, 2); + tab_float (t, 2, i * 3 + 3, 0, sst, 10, 2); + tab_float (t, 2, i * 3 + 2, 0, sst - ssa, 10, 2); + + + /* Degrees of freedom */ + tab_float (t, 3, i * 3 + 1, 0, df1, 4, 0); + tab_float (t, 3, i * 3 + 2, 0, df2, 4, 0); + tab_float (t, 3, i * 3 + 3, 0, totals->n - 1, 4, 0); + + /* 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.grp_data.mse, 8, 3); + + + { + const double F = msa/vars[i]->p.grp_data.mse ; + + /* The F value */ + tab_float (t, 5, i * 3 + 1, 0, F, 8, 3); + + /* The significance */ + tab_float (t, 6, i * 3 + 1, 0, gsl_cdf_fdist_Q(F,df1,df2), 8, 3); + } - const union value *val = case_data (&c, indep_var->fv); - - hsh_insert(ind_vals, (void *) val); + } - /* - if (! value_is_missing(val,v) ) - { - gs->n+=weight; - gs->sum+=weight * val->f; - gs->ssq+=weight * val->f * val->f; - } - */ - } - casereader_destroy (r); - n_groups = hsh_count(ind_vals); + tab_title (t, 0, _("ANOVA")); + tab_submit (t); } - /* Show the descriptives table */ static void show_descriptives(void) { int v; int n_cols =10; - int n_rows = n_vars * (n_groups + 1 )+ 2; - struct tab_table *t; + int row; + + const double confidence=0.95; + const double q = (1.0 - confidence) / 2.0; + + + int n_rows = 2 ; + + for ( v = 0 ; v < n_vars ; ++v ) + 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); tab_dim (t, tab_natural_dimensions); @@ -332,7 +398,7 @@ show_descriptives(void) tab_vline(t, TAL_0, 7, 0, 0); tab_hline(t, TAL_1, 6, 7, 1); - tab_joint_text (t, 6, 0, 7, 0, TAB_CENTER | TAT_TITLE, _("95% Confidence Interval for Mean")); + tab_joint_text (t, 6, 0, 7, 0, TAB_CENTER | TAT_TITLE | TAT_PRINTF, _("%g%% Confidence Interval for Mean"),confidence*100.0); tab_text (t, 6, 1, TAB_CENTER | TAT_TITLE, _("Lower Bound")); tab_text (t, 7, 1, TAB_CENTER | TAT_TITLE, _("Upper Bound")); @@ -341,43 +407,107 @@ show_descriptives(void) tab_text (t, 9, 1, TAB_CENTER | TAT_TITLE, _("Maximum")); - tab_title (t, 0, "Descriptives"); + tab_title (t, 0, _("Descriptives")); + row = 2; for ( v=0 ; v < n_vars ; ++v ) { + double T; + double std_error; + + struct hsh_iterator g; - union value *group_value; + struct group_statistics *gs; + struct group_statistics *totals = &vars[v]->p.grp_data.ugs; + int count = 0 ; char *s = (vars[v]->label) ? vars[v]->label : vars[v]->name; - tab_text (t, 0, v * ( n_groups + 1 ) + 2, TAB_LEFT | TAT_TITLE, s); + struct hsh_table *group_hash = vars[v]->p.grp_data.group_hash; + + + tab_text (t, 0, row, TAB_LEFT | TAT_TITLE, s); if ( v > 0) - tab_hline(t, TAL_1, 0, n_cols - 1 , v * (n_groups + 1) + 2); + tab_hline(t, TAL_1, 0, n_cols - 1 , row); - for (group_value = hsh_first (ind_vals,&g); - group_value != 0; - group_value = hsh_next(ind_vals,&g)) + for (gs = hsh_first (group_hash,&g); + gs != 0; + gs = hsh_next(group_hash,&g)) { - char *lab; - - lab = val_labs_find(indep_var->val_labs,*group_value); + const char *s = val_labs_find(indep_var->val_labs, gs->id ); - if ( lab ) - tab_text (t, 1, v * (n_groups + 1)+ count + 2, - TAB_LEFT | TAT_TITLE ,lab); + if ( s ) + tab_text (t, 1, row + count, + TAB_LEFT | TAT_TITLE ,s); + else if ( indep_var->width != 0 ) + tab_text (t, 1, row + count, + TAB_LEFT | TAT_TITLE, gs->id.s); else - tab_text (t, 1, v * (n_groups + 1) + count + 2, - TAB_LEFT | TAT_TITLE | TAT_PRINTF, "%g", group_value->f); + tab_text (t, 1, row + count, + TAB_LEFT | TAT_TITLE | TAT_PRINTF, "%g", gs->id.f); + + + /* Now fill in the numbers ... */ + + tab_float (t, 2, row + count, 0, gs->n, 8,0); + + tab_float (t, 3, row + count, 0, gs->mean,8,2); + tab_float (t, 4, row + count, 0, gs->std_dev,8,2); + + std_error = gs->std_dev/sqrt(gs->n) ; + tab_float (t, 5, row + count, 0, + 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 * std_error, 8, 2); + + tab_float(t, 7, row + count, 0, + gs->mean + T * std_error, 8, 2); + + /* Min and Max */ + + tab_float(t, 8, row + count, 0, gs->minimum, 8, 2); + tab_float(t, 9, row + count, 0, gs->maximum, 8, 2); + count++ ; } - tab_text (t, 1, v * (n_groups + 1)+ count + 2, + tab_text (t, 1, row + count, TAB_LEFT | TAT_TITLE ,_("Total")); + + tab_float (t, 2, row + count, 0, totals->n, 8,0); + + tab_float (t, 3, row + count, 0, totals->mean, 8,2); + + tab_float (t, 4, row + count, 0, totals->std_dev,8,2); + + std_error = totals->std_dev/sqrt(totals->n) ; + + 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 * std_error, 8, 2); + + tab_float(t, 7, row + count, 0, + 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.grp_data.n_groups + 1; } @@ -386,7 +516,6 @@ show_descriptives(void) } - /* Show the homogeneity table */ static void show_homogeneity(void) @@ -424,9 +553,23 @@ 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); @@ -437,10 +580,10 @@ show_homogeneity(void) /* Show the contrast coefficients table */ static void -show_contrast_coeffs(void) +show_contrast_coeffs(short *bad_contrast) { char *s; - int n_cols = 2 + n_groups; + int n_cols = 2 + ostensible_number_of_groups; int n_rows = 2 + cmd.sbc_contrast; struct hsh_iterator g; union value *group_value; @@ -490,13 +633,14 @@ show_contrast_coeffs(void) tab_joint_text (t, 2, 0, n_cols - 1, 0, TAB_CENTER | TAT_TITLE, s); - for (group_value = hsh_first (ind_vals,&g); + for (group_value = hsh_first (global_group_hash,&g); group_value != 0; - group_value = hsh_next(ind_vals,&g)) + group_value = hsh_next(global_group_hash,&g)) { int i; char *lab; + lab = val_labs_find(indep_var->val_labs,*group_value); if ( lab ) @@ -508,10 +652,15 @@ show_contrast_coeffs(void) for (i = 0 ; i < cmd.sbc_contrast ; ++i ) { + tab_text(t, 1, i + 2, TAB_CENTER | TAT_PRINTF, "%d", i + 1); - tab_text(t, count + 2, i + 2, TAB_RIGHT | TAT_PRINTF, "%g", - subc_list_double_at(&cmd.dl_contrast[i],count) - ); + + if ( bad_contrast[i] ) + tab_text(t, count + 2, i + 2, TAB_RIGHT, "?" ); + else + tab_text(t, count + 2, i + 2, TAB_RIGHT | TAT_PRINTF, "%g", + subc_list_double_at(&cmd.dl_contrast[i],count) + ); } count++ ; @@ -522,10 +671,9 @@ show_contrast_coeffs(void) } - /* Show the results of the contrast tests */ static void -show_contrast_tests(void) +show_contrast_tests(short *bad_contrast) { int v; int n_cols = 8; @@ -544,7 +692,6 @@ show_contrast_tests(void) 0, 0, n_cols - 1, n_rows - 1); - tab_box (t, -1,-1, TAL_0, TAL_0, @@ -569,26 +716,156 @@ show_contrast_tests(void) int i; int lines_per_variable = 2 * cmd.sbc_contrast; + tab_text (t, 0, (v * lines_per_variable) + 1, TAB_LEFT | TAT_TITLE, vars[v]->label?vars[v]->label:vars[v]->name); + + for ( i = 0 ; i < cmd.sbc_contrast ; ++i ) { - tab_text (t, 1, (v * lines_per_variable) + i*2 + 1, - TAB_LEFT | TAT_TITLE, - _("Assume equal variances")); + int ci; + double contrast_value = 0.0; + double coef_msq = 0.0; + 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 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) + } + */ - tab_text (t, 1, (v * lines_per_variable) + i*2 + 2, - TAB_LEFT | TAT_TITLE, - _("Does not assume equal")); + double df_denominator = 0.0; + double df_numerator = 0.0; + + if ( i == 0 ) + { + tab_text (t, 1, (v * lines_per_variable) + i + 1, + TAB_LEFT | TAT_TITLE, + _("Assume equal variances")); - tab_text (t, 2, (v * lines_per_variable) + i*2 + 1, + tab_text (t, 1, (v * lines_per_variable) + i + 1 + cmd.sbc_contrast, + TAB_LEFT | TAT_TITLE, + _("Does not assume equal")); + } + + tab_text (t, 2, (v * lines_per_variable) + i + 1, TAB_CENTER | TAT_TITLE | TAT_PRINTF, "%d",i+1); - tab_text (t, 2, (v * lines_per_variable) + i*2 + 2, + + tab_text (t, 2, (v * lines_per_variable) + i + 1 + cmd.sbc_contrast, TAB_CENTER | TAT_TITLE | TAT_PRINTF, "%d",i+1); + + if ( bad_contrast[i]) + continue; + + /* FIXME: Potential danger here. + We're ASSUMING THE array is in the order corresponding to the + hash order. */ + for (ci = 0, gs = hsh_first (group_hash,&g); + 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); + + tab_float (t, 3, (v * lines_per_variable) + i + 1 + + cmd.sbc_contrast, + TAB_RIGHT, contrast_value, 8,2); + + 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, std_error_contrast, + 8,3); + + T = fabs(contrast_value / std_error_contrast) ; + + /* T Statistic */ + + tab_float (t, 5, (v * lines_per_variable) + i + 1, + TAB_RIGHT, T, + 8,3); + + df = grp_data->ugs.n - grp_data->n_groups; + + /* Degrees of Freedom */ + tab_float (t, 6, (v * lines_per_variable) + i + 1, + TAB_RIGHT, df, + 8,0); + + + /* Significance TWO TAILED !!*/ + tab_float (t, 7, (v * lines_per_variable) + i + 1, + TAB_RIGHT, 2 * gsl_cdf_tdist_Q(T,df), + 8,3); + + + /* Now for the Variances NOT Equal case */ + + /* 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); + + + df = df_numerator / df_denominator; + + tab_float (t, 6, + (v * lines_per_variable) + i + 1 + cmd.sbc_contrast, + TAB_RIGHT, df, + 8,3); + + /* The Significance */ + + tab_float (t, 7, (v * lines_per_variable) + i + 1 + cmd.sbc_contrast, + TAB_RIGHT, 2 * gsl_cdf_tdist_Q(T,df), + 8,3); + + } if ( v > 0 ) @@ -598,3 +875,216 @@ show_contrast_tests(void) tab_submit (t); } + + +/* ONEWAY ANOVA Calculations */ + +static void postcalc ( struct cmd_oneway *cmd UNUSED ); + +static void precalc ( struct cmd_oneway *cmd UNUSED ); + + + +/* Pre calculations */ +static void +precalc ( struct cmd_oneway *cmd UNUSED ) +{ + int i=0; + + for(i=0; i< n_vars ; ++i) + { + 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.grp_data.group_hash = + hsh_create(4, + (hsh_compare_func *) compare_group, + (hsh_hash_func *) hash_group, + (hsh_free_func *) free_group, + (void *) indep_var->width ); + + + totals->sum=0; + totals->n=0; + totals->ssq=0; + totals->sum_diff=0; + totals->maximum = - DBL_MAX; + totals->minimum = DBL_MAX; + } +} + + +static void +run_oneway(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, + (hsh_compare_func *) compare_values, + (hsh_hash_func *) hash_value, + 0, + (void *) indep_var->width ); + precalc(cmd); + + for(r = casefile_get_reader (cf); + casereader_read (r, &c) ; + case_destroy (&c)) + { + int i; + + const double weight = + 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.grp_data.group_hash; + + struct group_statistics *gs; + + gs = hsh_find(group_hash, (void *) indep_val ); + + if ( ! gs ) + { + gs = (struct group_statistics *) + xmalloc (sizeof(struct group_statistics)); + + gs->id = *indep_val; + gs->sum=0; + gs->n=0; + gs->ssq=0; + gs->sum_diff=0; + gs->minimum = DBL_MAX; + gs->maximum = -DBL_MAX; + + hsh_insert ( group_hash, (void *) gs ); + } + + if (! value_is_missing(val,v) ) + { + struct group_statistics *totals = &vars[i]->p.grp_data.ugs; + + totals->n+=weight; + totals->sum+=weight * val->f; + totals->ssq+=weight * val->f * val->f; + + if ( val->f * weight < totals->minimum ) + totals->minimum = val->f * weight; + + if ( val->f * weight > totals->maximum ) + totals->maximum = val->f * weight; + + gs->n+=weight; + gs->sum+=weight * val->f; + gs->ssq+=weight * val->f * val->f; + + if ( val->f * weight < gs->minimum ) + gs->minimum = val->f * weight; + + if ( val->f * weight > gs->maximum ) + gs->maximum = val->f * weight; + } + + vars[i]->p.grp_data.n_groups = hsh_count ( group_hash ); + } + + } + casereader_destroy (r); + + 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); + + + output_oneway(); + + +} + + +/* Post calculations for the ONEWAY command */ +void +postcalc ( struct cmd_oneway *cmd UNUSED ) +{ + int i=0; + + + for(i = 0; i < n_vars ; ++i) + { + 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; + + for (gs = hsh_first (group_hash,&g); + gs != 0; + gs = hsh_next(group_hash,&g)) + { + gs->mean=gs->sum / gs->n; + gs->s_std_dev= sqrt( + ( (gs->ssq / gs->n ) - gs->mean * gs->mean ) + ) ; + + gs->std_dev= sqrt( + gs->n/(gs->n-1) * + ( (gs->ssq / gs->n ) - gs->mean * gs->mean ) + ) ; + + gs->se_mean = gs->std_dev / sqrt(gs->n); + gs->mean_diff= gs->sum_diff / gs->n; + + } + + + + totals->mean = totals->sum / totals->n; + totals->std_dev= sqrt( + totals->n/(totals->n-1) * + ( (totals->ssq / totals->n ) - totals->mean * totals->mean ) + ) ; + + totals->se_mean = totals->std_dev / sqrt(totals->n); + + } +}