X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=src%2Flanguage%2Fstats%2Foneway.c;h=0fc6e1fcb27604447832d4d1080b10f363aa03a6;hb=e2da62d735c597afeef2e0e9b36e5a4a83d7da94;hp=f481c69dc13c765cfd55cbc346ca2089a15df7f2;hpb=af097b7430b4080eec4399df6898a9e97d17d0b4;p=pspp diff --git a/src/language/stats/oneway.c b/src/language/stats/oneway.c index f481c69dc1..0fc6e1fcb2 100644 --- a/src/language/stats/oneway.c +++ b/src/language/stats/oneway.c @@ -1,5 +1,5 @@ /* PSPP - a program for statistical analysis. - Copyright (C) 1997-9, 2000, 2007, 2009, 2010 Free Software Foundation, Inc. + Copyright (C) 1997-9, 2000, 2007, 2009, 2010, 2011, 2012, 2013 Free Software Foundation, Inc. This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by @@ -16,44 +16,67 @@ #include -#include -#include -#include - -#include -#include -#include +#include #include -#include +#include -#include +#include "data/case.h" +#include "data/casegrouper.h" +#include "data/casereader.h" +#include "data/dataset.h" +#include "data/dictionary.h" +#include "data/format.h" +#include "data/value.h" +#include "language/command.h" +#include "language/dictionary/split-file.h" +#include "language/lexer/lexer.h" +#include "language/lexer/value-parser.h" +#include "language/lexer/variable-parser.h" +#include "libpspp/ll.h" +#include "libpspp/message.h" +#include "libpspp/misc.h" +#include "libpspp/taint.h" +#include "linreg/sweep.h" +#include "tukey/tukey.h" +#include "math/categoricals.h" +#include "math/interaction.h" +#include "math/covariance.h" +#include "math/levene.h" +#include "math/moments.h" +#include "output/tab.h" -#include -#include -#include -#include +#include "gettext.h" +#define _(msgid) gettext (msgid) +#define N_(msgid) msgid -#include -#include -#include +/* Workspace variable for each dependent variable */ +struct per_var_ws +{ + struct interaction *iact; + struct categoricals *cat; + struct covariance *cov; + struct levene *nl; -#include -#include -#include -#include -#include -#include + double n; -#include + double sst; + double sse; + double ssa; -#include -#include -#include + int n_groups; -#include + double mse; +}; -#include "gettext.h" -#define _(msgid) gettext (msgid) +/* Per category data */ +struct descriptive_data +{ + const struct variable *var; + struct moments1 *mom; + + double minimum; + double maximum; +}; enum missing_type { @@ -78,8 +101,28 @@ struct contrasts_node { struct ll ll; struct ll_list coefficient_list; +}; + + +struct oneway_spec; + +typedef double df_func (const struct per_var_ws *pvw, const struct moments1 *mom_i, const struct moments1 *mom_j); +typedef double ts_func (int k, const struct moments1 *mom_i, const struct moments1 *mom_j, double std_err); +typedef double p1tail_func (double ts, double df1, double df2); + +typedef double pinv_func (double std_err, double alpha, double df, int k, const struct moments1 *mom_i, const struct moments1 *mom_j); + + +struct posthoc +{ + const char *syntax; + const char *label; + + df_func *dff; + ts_func *tsf; + p1tail_func *p1f; - bool bad_count; /* True if the number of coefficients does not equal the number of groups */ + pinv_func *pinv; }; struct oneway_spec @@ -100,23 +143,227 @@ struct oneway_spec /* The weight variable */ const struct variable *wv; + /* The confidence level for multiple comparisons */ + double alpha; + + int *posthoc; + int n_posthoc; }; +static double +df_common (const struct per_var_ws *pvw, const struct moments1 *mom_i UNUSED, const struct moments1 *mom_j UNUSED) +{ + return pvw->n - pvw->n_groups; +} -/* Workspace variable for each dependent variable */ -struct per_var_ws +static double +df_individual (const struct per_var_ws *pvw UNUSED, const struct moments1 *mom_i, const struct moments1 *mom_j) { - struct covariance *cov; + double n_i, var_i; + double n_j, var_j; + double nom,denom; - double sst; - double sse; - double ssa; + moments1_calculate (mom_i, &n_i, NULL, &var_i, 0, 0); + moments1_calculate (mom_j, &n_j, NULL, &var_j, 0, 0); + + if ( n_i <= 1.0 || n_j <= 1.0) + return SYSMIS; - int n_groups; + nom = pow2 (var_i/n_i + var_j/n_j); + denom = pow2 (var_i/n_i) / (n_i - 1) + pow2 (var_j/n_j) / (n_j - 1); + + return nom / denom; +} + +static double lsd_pinv (double std_err, double alpha, double df, int k UNUSED, const struct moments1 *mom_i UNUSED, const struct moments1 *mom_j UNUSED) +{ + return std_err * gsl_cdf_tdist_Pinv (1.0 - alpha / 2.0, df); +} + +static double bonferroni_pinv (double std_err, double alpha, double df, int k, const struct moments1 *mom_i UNUSED, const struct moments1 *mom_j UNUSED) +{ + const int m = k * (k - 1) / 2; + return std_err * gsl_cdf_tdist_Pinv (1.0 - alpha / (2.0 * m), df); +} + +static double sidak_pinv (double std_err, double alpha, double df, int k, const struct moments1 *mom_i UNUSED, const struct moments1 *mom_j UNUSED) +{ + const double m = k * (k - 1) / 2; + double lp = 1.0 - exp (log (1.0 - alpha) / m ) ; + return std_err * gsl_cdf_tdist_Pinv (1.0 - lp / 2.0, df); +} + +static double tukey_pinv (double std_err, double alpha, double df, int k, const struct moments1 *mom_i UNUSED, const struct moments1 *mom_j UNUSED) +{ + if ( k < 2 || df < 2) + return SYSMIS; + + return std_err / sqrt (2.0) * qtukey (1 - alpha, 1.0, k, df, 1, 0); +} + +static double scheffe_pinv (double std_err, double alpha, double df, int k, const struct moments1 *mom_i UNUSED, const struct moments1 *mom_j UNUSED) +{ + double x = (k - 1) * gsl_cdf_fdist_Pinv (1.0 - alpha, k - 1, df); + return std_err * sqrt (x); +} + +static double gh_pinv (double std_err UNUSED, double alpha, double df, int k, const struct moments1 *mom_i, const struct moments1 *mom_j) +{ + double n_i, mean_i, var_i; + double n_j, mean_j, var_j; + double m; + + moments1_calculate (mom_i, &n_i, &mean_i, &var_i, 0, 0); + moments1_calculate (mom_j, &n_j, &mean_j, &var_j, 0, 0); + + m = sqrt ((var_i/n_i + var_j/n_j) / 2.0); + + if ( k < 2 || df < 2) + return SYSMIS; + + return m * qtukey (1 - alpha, 1.0, k, df, 1, 0); +} + + +static double +multiple_comparison_sig (double std_err, + const struct per_var_ws *pvw, + const struct descriptive_data *dd_i, const struct descriptive_data *dd_j, + const struct posthoc *ph) +{ + int k = pvw->n_groups; + double df = ph->dff (pvw, dd_i->mom, dd_j->mom); + double ts = ph->tsf (k, dd_i->mom, dd_j->mom, std_err); + if ( df == SYSMIS) + return SYSMIS; + return ph->p1f (ts, k - 1, df); +} + +static double +mc_half_range (const struct oneway_spec *cmd, const struct per_var_ws *pvw, double std_err, const struct descriptive_data *dd_i, const struct descriptive_data *dd_j, const struct posthoc *ph) +{ + int k = pvw->n_groups; + double df = ph->dff (pvw, dd_i->mom, dd_j->mom); + if ( df == SYSMIS) + return SYSMIS; + + return ph->pinv (std_err, cmd->alpha, df, k, dd_i->mom, dd_j->mom); +} + +static double tukey_1tailsig (double ts, double df1, double df2) +{ + double twotailedsig; + + if (df2 < 2 || df1 < 1) + return SYSMIS; + + twotailedsig = 1.0 - ptukey (ts, 1.0, df1 + 1, df2, 1, 0); + + return twotailedsig / 2.0; +} + +static double lsd_1tailsig (double ts, double df1 UNUSED, double df2) +{ + return ts < 0 ? gsl_cdf_tdist_P (ts, df2) : gsl_cdf_tdist_Q (ts, df2); +} + +static double sidak_1tailsig (double ts, double df1, double df2) +{ + double ex = (df1 + 1.0) * df1 / 2.0; + double lsd_sig = 2 * lsd_1tailsig (ts, df1, df2); + + return 0.5 * (1.0 - pow (1.0 - lsd_sig, ex)); +} + +static double bonferroni_1tailsig (double ts, double df1, double df2) +{ + const int m = (df1 + 1) * df1 / 2; + + double p = ts < 0 ? gsl_cdf_tdist_P (ts, df2) : gsl_cdf_tdist_Q (ts, df2); + p *= m; + + return p > 0.5 ? 0.5 : p; +} + +static double scheffe_1tailsig (double ts, double df1, double df2) +{ + return 0.5 * gsl_cdf_fdist_Q (ts, df1, df2); +} + + +static double tukey_test_stat (int k UNUSED, const struct moments1 *mom_i, const struct moments1 *mom_j, double std_err) +{ + double ts; + double n_i, mean_i, var_i; + double n_j, mean_j, var_j; + + moments1_calculate (mom_i, &n_i, &mean_i, &var_i, 0, 0); + moments1_calculate (mom_j, &n_j, &mean_j, &var_j, 0, 0); + + ts = (mean_i - mean_j) / std_err; + ts = fabs (ts) * sqrt (2.0); + + return ts; +} + +static double lsd_test_stat (int k UNUSED, const struct moments1 *mom_i, const struct moments1 *mom_j, double std_err) +{ + double n_i, mean_i, var_i; + double n_j, mean_j, var_j; + + moments1_calculate (mom_i, &n_i, &mean_i, &var_i, 0, 0); + moments1_calculate (mom_j, &n_j, &mean_j, &var_j, 0, 0); + + return (mean_i - mean_j) / std_err; +} + +static double scheffe_test_stat (int k, const struct moments1 *mom_i, const struct moments1 *mom_j, double std_err) +{ + double t; + double n_i, mean_i, var_i; + double n_j, mean_j, var_j; + + moments1_calculate (mom_i, &n_i, &mean_i, &var_i, 0, 0); + moments1_calculate (mom_j, &n_j, &mean_j, &var_j, 0, 0); + + t = (mean_i - mean_j) / std_err; + t = pow2 (t); + t /= k - 1; + + return t; +} + +static double gh_test_stat (int k UNUSED, const struct moments1 *mom_i, const struct moments1 *mom_j, double std_err UNUSED) +{ + double ts; + double thing; + double n_i, mean_i, var_i; + double n_j, mean_j, var_j; + + moments1_calculate (mom_i, &n_i, &mean_i, &var_i, 0, 0); + moments1_calculate (mom_j, &n_j, &mean_j, &var_j, 0, 0); + + thing = var_i / n_i + var_j / n_j; + thing /= 2.0; + thing = sqrt (thing); + + ts = (mean_i - mean_j) / thing; + + return fabs (ts); +} + + + +static const struct posthoc ph_tests [] = + { + { "LSD", N_("LSD"), df_common, lsd_test_stat, lsd_1tailsig, lsd_pinv}, + { "TUKEY", N_("Tukey HSD"), df_common, tukey_test_stat, tukey_1tailsig, tukey_pinv}, + { "BONFERRONI", N_("Bonferroni"), df_common, lsd_test_stat, bonferroni_1tailsig, bonferroni_pinv}, + { "SCHEFFE", N_("Scheffé"), df_common, scheffe_test_stat, scheffe_1tailsig, scheffe_pinv}, + { "GH", N_("Games-Howell"), df_individual, gh_test_stat, tukey_1tailsig, gh_pinv}, + { "SIDAK", N_("Šidák"), df_common, lsd_test_stat, sidak_1tailsig, sidak_pinv} + }; - double cc; - double mse; -}; struct oneway_workspace { @@ -124,10 +371,6 @@ struct oneway_workspace missing values are disregarded */ int actual_number_of_groups; - /* A hash table containing all the distinct values of the independent - variable */ - struct hsh_table *group_hash; - struct per_var_ws *vws; /* An array of descriptive data. One for each dependent variable */ @@ -142,6 +385,37 @@ static void show_homogeneity (const struct oneway_spec *, const struct oneway_wo static void output_oneway (const struct oneway_spec *, struct oneway_workspace *ws); static void run_oneway (const struct oneway_spec *cmd, struct casereader *input, const struct dataset *ds); + +static void +destroy_coeff_list (struct contrasts_node *coeff_list) +{ + struct coeff_node *cn = NULL; + struct coeff_node *cnx = NULL; + struct ll_list *cl = &coeff_list->coefficient_list; + + ll_for_each_safe (cn, cnx, struct coeff_node, ll, cl) + { + free (cn); + } + + free (coeff_list); +} + +static void +oneway_cleanup (struct oneway_spec *cmd) +{ + struct contrasts_node *coeff_list = NULL; + struct contrasts_node *coeff_next = NULL; + ll_for_each_safe (coeff_list, coeff_next, struct contrasts_node, ll, &cmd->contrast_list) + { + destroy_coeff_list (coeff_list); + } + + free (cmd->posthoc); +} + + + int cmd_oneway (struct lexer *lexer, struct dataset *ds) { @@ -154,17 +428,20 @@ cmd_oneway (struct lexer *lexer, struct dataset *ds) oneway.missing_type = MISS_ANALYSIS; oneway.exclude = MV_ANY; oneway.wv = dict_get_weight (dict); + oneway.alpha = 0.05; + oneway.posthoc = NULL; + oneway.n_posthoc = 0; ll_init (&oneway.contrast_list); - if ( lex_match (lexer, '/')) + if ( lex_match (lexer, T_SLASH)) { if (!lex_force_match_id (lexer, "VARIABLES")) { goto error; } - lex_match (lexer, '='); + lex_match (lexer, T_EQUALS); } if (!parse_variables_const (lexer, dict, @@ -176,14 +453,14 @@ cmd_oneway (struct lexer *lexer, struct dataset *ds) oneway.indep_var = parse_variable_const (lexer, dict); - while (lex_token (lexer) != '.') + while (lex_token (lexer) != T_ENDCMD) { - lex_match (lexer, '/'); + lex_match (lexer, T_SLASH); if (lex_match_id (lexer, "STATISTICS")) { - lex_match (lexer, '='); - while (lex_token (lexer) != '.' && lex_token (lexer) != '/') + lex_match (lexer, T_EQUALS); + while (lex_token (lexer) != T_ENDCMD && lex_token (lexer) != T_SLASH) { if (lex_match_id (lexer, "DESCRIPTIVES")) { @@ -200,16 +477,55 @@ cmd_oneway (struct lexer *lexer, struct dataset *ds) } } } + else if (lex_match_id (lexer, "POSTHOC")) + { + lex_match (lexer, T_EQUALS); + while (lex_token (lexer) != T_ENDCMD && lex_token (lexer) != T_SLASH) + { + int p; + bool method = false; + for (p = 0 ; p < sizeof (ph_tests) / sizeof (struct posthoc); ++p) + { + if (lex_match_id (lexer, ph_tests[p].syntax)) + { + oneway.n_posthoc++; + oneway.posthoc = xrealloc (oneway.posthoc, sizeof (*oneway.posthoc) * oneway.n_posthoc); + oneway.posthoc[oneway.n_posthoc - 1] = p; + method = true; + break; + } + } + if ( method == false) + { + if (lex_match_id (lexer, "ALPHA")) + { + if ( !lex_force_match (lexer, T_LPAREN)) + goto error; + lex_force_num (lexer); + oneway.alpha = lex_number (lexer); + lex_get (lexer); + if ( !lex_force_match (lexer, T_RPAREN)) + goto error; + } + else + { + msg (SE, _("The post hoc analysis method %s is not supported."), lex_tokcstr (lexer)); + lex_error (lexer, NULL); + goto error; + } + } + } + } else if (lex_match_id (lexer, "CONTRAST")) { struct contrasts_node *cl = xzalloc (sizeof *cl); struct ll_list *coefficient_list = &cl->coefficient_list; - lex_match (lexer, '='); + lex_match (lexer, T_EQUALS); ll_init (coefficient_list); - while (lex_token (lexer) != '.' && lex_token (lexer) != '/') + while (lex_token (lexer) != T_ENDCMD && lex_token (lexer) != T_SLASH) { if ( lex_is_number (lexer)) { @@ -221,6 +537,7 @@ cmd_oneway (struct lexer *lexer, struct dataset *ds) } else { + destroy_coeff_list (cl); lex_error (lexer, NULL); goto error; } @@ -230,8 +547,8 @@ cmd_oneway (struct lexer *lexer, struct dataset *ds) } else if (lex_match_id (lexer, "MISSING")) { - lex_match (lexer, '='); - while (lex_token (lexer) != '.' && lex_token (lexer) != '/') + lex_match (lexer, T_EQUALS); + while (lex_token (lexer) != T_ENDCMD && lex_token (lexer) != T_SLASH) { if (lex_match_id (lexer, "INCLUDE")) { @@ -276,10 +593,12 @@ cmd_oneway (struct lexer *lexer, struct dataset *ds) ok = proc_commit (ds) && ok; } + oneway_cleanup (&oneway); free (oneway.vars); return CMD_SUCCESS; error: + oneway_cleanup (&oneway); free (oneway.vars); return CMD_FAILURE; } @@ -287,41 +606,6 @@ cmd_oneway (struct lexer *lexer, struct dataset *ds) -static int -compare_double_3way (const void *a_, const void *b_, const void *aux UNUSED) -{ - const double *a = a_; - const double *b = b_; - return *a < *b ? -1 : *a > *b; -} - -static unsigned -do_hash_double (const void *value_, const void *aux UNUSED) -{ - const double *value = value_; - return hash_double (*value, 0); -} - -static void -free_double (void *value_, const void *aux UNUSED) -{ - double *value = value_; - free (value); -} - - - -static void postcalc (const struct oneway_spec *cmd); -static void precalc (const struct oneway_spec *cmd); - -struct descriptive_data -{ - const struct variable *var; - struct moments1 *mom; - - double minimum; - double maximum; -}; static struct descriptive_data * dd_create (const struct variable *var) @@ -336,9 +620,15 @@ dd_create (const struct variable *var) return dd; } +static void +dd_destroy (struct descriptive_data *dd) +{ + moments1_destroy (dd->mom); + free (dd); +} static void * -makeit (void *aux1, void *aux2 UNUSED) +makeit (const void *aux1, void *aux2 UNUSED) { const struct variable *var = aux1; @@ -348,10 +638,17 @@ makeit (void *aux1, void *aux2 UNUSED) } static void -updateit (void *user_data, const struct variable *wv, - const struct variable *catvar UNUSED, - const struct ccase *c, - void *aux1, void *aux2) +killit (const void *aux1 UNUSED, void *aux2 UNUSED, void *user_data) +{ + struct descriptive_data *dd = user_data; + + dd_destroy (dd); +} + + +static void +updateit (const void *aux1, void *aux2, void *user_data, + const struct ccase *c, double weight) { struct descriptive_data *dd = user_data; @@ -361,16 +658,12 @@ updateit (void *user_data, const struct variable *wv, struct descriptive_data *dd_total = aux2; - double weight = 1.0; - if (wv) - weight = case_data (c, wv)->f; - moments1_add (dd->mom, valx->f, weight); - if (valx->f * weight < dd->minimum) - dd->minimum = valx->f * weight; + if (valx->f < dd->minimum) + dd->minimum = valx->f; - if (valx->f * weight > dd->maximum) - dd->maximum = valx->f * weight; + if (valx->f > dd->maximum) + dd->maximum = valx->f; { const struct variable *var = dd_total->var; @@ -380,11 +673,11 @@ updateit (void *user_data, const struct variable *wv, val->f, weight); - if (val->f * weight < dd_total->minimum) - dd_total->minimum = val->f * weight; + if (val->f < dd_total->minimum) + dd_total->minimum = val->f; - if (val->f * weight > dd_total->maximum) - dd_total->maximum = val->f * weight; + if (val->f > dd_total->maximum) + dd_total->maximum = val->f; } } @@ -400,51 +693,48 @@ run_oneway (const struct oneway_spec *cmd, struct ccase *c; struct oneway_workspace ws; - - { - ws.vws = xmalloc (cmd->n_vars * sizeof (*ws.vws)); - ws.dd_total = xmalloc (sizeof (struct descriptive_data) * cmd->n_vars); + ws.actual_number_of_groups = 0; + ws.vws = xzalloc (cmd->n_vars * sizeof (*ws.vws)); + ws.dd_total = xmalloc (sizeof (struct descriptive_data) * cmd->n_vars); - for (v = 0 ; v < cmd->n_vars; ++v) - { - ws.dd_total[v] = dd_create (cmd->vars[v]); - } - } + for (v = 0 ; v < cmd->n_vars; ++v) + ws.dd_total[v] = dd_create (cmd->vars[v]); for (v = 0; v < cmd->n_vars; ++v) { - struct categoricals *cats = categoricals_create (&cmd->indep_var, 1, - cmd->wv, cmd->exclude, - makeit, - updateit, - cmd->vars[v], ws.dd_total[v]); + struct payload payload; + payload.create = makeit; + payload.update = updateit; + payload.calculate = NULL; + payload.destroy = killit; + + ws.vws[v].iact = interaction_create (cmd->indep_var); + ws.vws[v].cat = categoricals_create (&ws.vws[v].iact, 1, cmd->wv, + cmd->exclude, cmd->exclude); + + categoricals_set_payload (ws.vws[v].cat, &payload, + CONST_CAST (struct variable *, cmd->vars[v]), + ws.dd_total[v]); + ws.vws[v].cov = covariance_2pass_create (1, &cmd->vars[v], - cats, + ws.vws[v].cat, cmd->wv, cmd->exclude); - ws.vws[v].cc = 0; + ws.vws[v].nl = levene_create (var_get_width (cmd->indep_var), NULL); } c = casereader_peek (input, 0); if (c == NULL) { casereader_destroy (input); - return; + goto finish; } output_split_file_values (ds, c); case_unref (c); taint = taint_clone (casereader_get_taint (input)); - ws.group_hash = hsh_create (4, - compare_double_3way, - do_hash_double, - free_double, - cmd->indep_var); - - precalc (cmd); - input = casereader_create_filter_missing (input, &cmd->indep_var, 1, cmd->exclude, NULL, NULL); if (cmd->missing_type == MISS_LISTWISE) @@ -453,210 +743,150 @@ run_oneway (const struct oneway_spec *cmd, input = casereader_create_filter_weight (input, dict, NULL, NULL); reader = casereader_clone (input); - for (; (c = casereader_read (reader)) != NULL; case_unref (c)) { - size_t i; - - const double weight = dict_get_case_weight (dict, c, NULL); - - const union value *indep_val = case_data (c, cmd->indep_var); - void **p = hsh_probe (ws.group_hash, &indep_val->f); - if (*p == NULL) - { - double *value = *p = xmalloc (sizeof *value); - *value = indep_val->f; - } + int i; + double w = dict_get_case_weight (dict, c, NULL); for (i = 0; i < cmd->n_vars; ++i) { - { - struct per_var_ws *pvw = &ws.vws[i]; - - pvw->cc += weight; - covariance_accumulate_pass1 (pvw->cov, c); - } - + struct per_var_ws *pvw = &ws.vws[i]; const struct variable *v = cmd->vars[i]; - const union value *val = case_data (c, v); - struct group_proc *gp = group_proc_get (cmd->vars[i]); - struct hsh_table *group_hash = gp->group_hash; - - struct group_statistics *gs; - - gs = hsh_find (group_hash, indep_val ); - - if ( ! gs ) + if ( MISS_ANALYSIS == cmd->missing_type) { - gs = xmalloc (sizeof *gs); - 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, gs ); + if ( var_is_value_missing (v, val, cmd->exclude)) + continue; } - if (!var_is_value_missing (v, val, cmd->exclude)) - { - struct group_statistics *totals = &gp->ugs; - - totals->n += weight; - totals->sum += weight * val->f; - totals->ssq += weight * pow2 (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 * pow2 (val->f); + covariance_accumulate_pass1 (pvw->cov, c); + levene_pass_one (pvw->nl, val->f, w, case_data (c, cmd->indep_var)); + } + } + casereader_destroy (reader); - if ( val->f * weight < gs->minimum ) - gs->minimum = val->f * weight; + reader = casereader_clone (input); + for ( ; (c = casereader_read (reader) ); case_unref (c)) + { + int i; + double w = dict_get_case_weight (dict, c, NULL); + for (i = 0; i < cmd->n_vars; ++i) + { + struct per_var_ws *pvw = &ws.vws[i]; + const struct variable *v = cmd->vars[i]; + const union value *val = case_data (c, v); - if ( val->f * weight > gs->maximum ) - gs->maximum = val->f * weight; + if ( MISS_ANALYSIS == cmd->missing_type) + { + if ( var_is_value_missing (v, val, cmd->exclude)) + continue; } - gp->n_groups = hsh_count (group_hash ); + covariance_accumulate_pass2 (pvw->cov, c); + levene_pass_two (pvw->nl, val->f, w, case_data (c, cmd->indep_var)); } - } casereader_destroy (reader); + reader = casereader_clone (input); for ( ; (c = casereader_read (reader) ); case_unref (c)) { int i; + double w = dict_get_case_weight (dict, c, NULL); + for (i = 0; i < cmd->n_vars; ++i) { struct per_var_ws *pvw = &ws.vws[i]; - covariance_accumulate_pass2 (pvw->cov, c); + const struct variable *v = cmd->vars[i]; + const union value *val = case_data (c, v); + + if ( MISS_ANALYSIS == cmd->missing_type) + { + if ( var_is_value_missing (v, val, cmd->exclude)) + continue; + } + + levene_pass_three (pvw->nl, val->f, w, case_data (c, cmd->indep_var)); } } casereader_destroy (reader); + for (v = 0; v < cmd->n_vars; ++v) { + const gsl_matrix *ucm; + gsl_matrix *cm; struct per_var_ws *pvw = &ws.vws[v]; - gsl_matrix *cm = covariance_calculate_unnormalized (pvw->cov); const struct categoricals *cats = covariance_get_categoricals (pvw->cov); + const bool ok = categoricals_sane (cats); + + if ( ! ok) + { + msg (MW, + _("Dependent variable %s has no non-missing values. No analysis for this variable will be done."), + var_get_name (cmd->vars[v])); + continue; + } + + ucm = covariance_calculate_unnormalized (pvw->cov); + + cm = gsl_matrix_alloc (ucm->size1, ucm->size2); + gsl_matrix_memcpy (cm, ucm); + + moments1_calculate (ws.dd_total[v]->mom, &pvw->n, NULL, NULL, NULL, NULL); pvw->sst = gsl_matrix_get (cm, 0, 0); reg_sweep (cm, 0); pvw->sse = gsl_matrix_get (cm, 0, 0); + gsl_matrix_free (cm); pvw->ssa = pvw->sst - pvw->sse; - pvw->n_groups = categoricals_total (cats); + pvw->n_groups = categoricals_n_total (cats); - pvw->mse = (pvw->sst - pvw->ssa) / (pvw->cc - pvw->n_groups); + pvw->mse = (pvw->sst - pvw->ssa) / (pvw->n - pvw->n_groups); } - postcalc (cmd); - for (v = 0; v < cmd->n_vars; ++v) { - struct categoricals *cats = covariance_get_categoricals (ws.vws[v].cov); + const struct categoricals *cats = covariance_get_categoricals (ws.vws[v].cov); - categoricals_done (cats); - } + if ( ! categoricals_is_complete (cats)) + { + continue; + } - if ( cmd->stats & STATS_HOMOGENEITY ) - levene (dict, casereader_clone (input), cmd->indep_var, - cmd->n_vars, cmd->vars, cmd->exclude); + if (categoricals_n_total (cats) > ws.actual_number_of_groups) + ws.actual_number_of_groups = categoricals_n_total (cats); + } casereader_destroy (input); - ws.actual_number_of_groups = hsh_count (ws.group_hash); - if (!taint_has_tainted_successor (taint)) output_oneway (cmd, &ws); taint_destroy (taint); -} -/* Pre calculations */ -static void -precalc (const struct oneway_spec *cmd) -{ - size_t i = 0; + finish: - for (i = 0; i < cmd->n_vars; ++i) + for (v = 0; v < cmd->n_vars; ++v) { - struct group_proc *gp = group_proc_get (cmd->vars[i]); - struct group_statistics *totals = &gp->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 */ - - gp->group_hash = hsh_create (4, compare_group, hash_group, - (hsh_free_func *) free_group, - cmd->indep_var); - - totals->sum = 0; - totals->n = 0; - totals->ssq = 0; - totals->sum_diff = 0; - totals->maximum = -DBL_MAX; - totals->minimum = DBL_MAX; + covariance_destroy (ws.vws[v].cov); + levene_destroy (ws.vws[v].nl); + dd_destroy (ws.dd_total[v]); + interaction_destroy (ws.vws[v].iact); } -} - -/* Post calculations for the ONEWAY command */ -static void -postcalc (const struct oneway_spec *cmd) -{ - size_t i = 0; - for (i = 0; i < cmd->n_vars; ++i) - { - struct group_proc *gp = group_proc_get (cmd->vars[i]); - struct hsh_table *group_hash = gp->group_hash; - struct group_statistics *totals = &gp->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 - pow2 (gs->mean)); - - gs->std_dev = sqrt ( - gs->n / (gs->n - 1) * - ( gs->ssq / gs->n - pow2 (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 - pow2 (totals->mean)) - ); - - totals->se_mean = totals->std_dev / sqrt (totals->n); - } + free (ws.vws); + free (ws.dd_total); } static void show_contrast_coeffs (const struct oneway_spec *cmd, const struct oneway_workspace *ws); static void show_contrast_tests (const struct oneway_spec *cmd, const struct oneway_workspace *ws); +static void show_comparisons (const struct oneway_spec *cmd, const struct oneway_workspace *ws, int depvar); static void output_oneway (const struct oneway_spec *cmd, struct oneway_workspace *ws) @@ -665,7 +895,8 @@ output_oneway (const struct oneway_spec *cmd, struct oneway_workspace *ws) /* Check the sanity of the given contrast values */ struct contrasts_node *coeff_list = NULL; - ll_for_each (coeff_list, struct contrasts_node, ll, &cmd->contrast_list) + struct contrasts_node *coeff_next = NULL; + ll_for_each_safe (coeff_list, coeff_next, struct contrasts_node, ll, &cmd->contrast_list) { struct coeff_node *cn = NULL; double sum = 0; @@ -675,8 +906,11 @@ output_oneway (const struct oneway_spec *cmd, struct oneway_workspace *ws) if (ll_count (cl) != ws->actual_number_of_groups) { msg (SW, - _("Number of contrast coefficients must equal the number of groups")); - coeff_list->bad_count = true; + _("In contrast list %zu, the number of coefficients (%zu) does not equal the number of groups (%d). This contrast list will be ignored."), + i, ll_count (cl), ws->actual_number_of_groups); + + ll_remove (&coeff_list->ll); + destroy_coeff_list (coeff_list); continue; } @@ -695,23 +929,23 @@ output_oneway (const struct oneway_spec *cmd, struct oneway_workspace *ws) show_anova_table (cmd, ws); - if (ll_count (&cmd->contrast_list) > 0) { show_contrast_coeffs (cmd, ws); show_contrast_tests (cmd, ws); } - - /* Clean up */ - for (i = 0; i < cmd->n_vars; ++i ) + if ( cmd->posthoc ) { - struct hsh_table *group_hash = group_proc_get (cmd->vars[i])->group_hash; + int v; + for (v = 0 ; v < cmd->n_vars; ++v) + { + const struct categoricals *cats = covariance_get_categoricals (ws->vws[v].cov); - hsh_destroy (group_hash); + if ( categoricals_is_complete (cats)) + show_comparisons (cmd, ws, v); + } } - - hsh_destroy (ws->group_hash); } @@ -746,12 +980,17 @@ show_anova_table (const struct oneway_spec *cmd, const struct oneway_workspace * for (i = 0; i < cmd->n_vars; ++i) { + double n; + double df1, df2; + double msa; + const char *s = var_to_string (cmd->vars[i]); const struct per_var_ws *pvw = &ws->vws[i]; - const double df1 = pvw->n_groups - 1; - const double df2 = pvw->cc - pvw->n_groups; - const double msa = pvw->ssa / df1; - const char *s = var_to_string (cmd->vars[i]); + moments1_calculate (ws->dd_total[i]->mom, &n, NULL, NULL, NULL, NULL); + + df1 = pvw->n_groups - 1; + df2 = n - pvw->n_groups; + msa = pvw->ssa / df1; 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")); @@ -771,7 +1010,7 @@ show_anova_table (const struct oneway_spec *cmd, const struct oneway_workspace * /* Degrees of freedom */ tab_fixed (t, 3, i * 3 + 1, 0, df1, 4, 0); tab_fixed (t, 3, i * 3 + 2, 0, df2, 4, 0); - tab_fixed (t, 3, i * 3 + 3, 0, pvw->cc - 1, 4, 0); + tab_fixed (t, 3, i * 3 + 3, 0, n - 1, 4, 0); /* Mean Squares */ tab_double (t, 4, i * 3 + 1, TAB_RIGHT, msa, NULL); @@ -861,7 +1100,7 @@ show_descriptives (const struct oneway_spec *cmd, const struct oneway_workspace if ( v > 0) tab_hline (t, TAL_1, 0, n_cols - 1, row); - for (count = 0; count < categoricals_total (cats); ++count) + for (count = 0; count < categoricals_n_total (cats); ++count) { double T; double n, mean, variance; @@ -869,8 +1108,8 @@ show_descriptives (const struct oneway_spec *cmd, const struct oneway_workspace struct string vstr; - const union value *gval = categoricals_get_value_by_subscript (cats, count); - const struct descriptive_data *dd = categoricals_get_user_data_by_subscript (cats, count); + const struct ccase *gcc = categoricals_get_case_by_category (cats, count); + const struct descriptive_data *dd = categoricals_get_user_data_by_category (cats, count); moments1_calculate (dd->mom, &n, &mean, &variance, NULL, NULL); @@ -879,7 +1118,7 @@ show_descriptives (const struct oneway_spec *cmd, const struct oneway_workspace ds_init_empty (&vstr); - var_append_value_name (cmd->indep_var, gval, &vstr); + var_append_value_name (cmd->indep_var, case_data (gcc, cmd->indep_var), &vstr); tab_text (t, 1, row + count, TAB_LEFT | TAT_TITLE, @@ -889,7 +1128,7 @@ show_descriptives (const struct oneway_spec *cmd, const struct oneway_workspace /* Now fill in the numbers ... */ - tab_fixed (t, 2, row + count, 0, n, 8, 0); + tab_double (t, 2, row + count, 0, n, wfmt); tab_double (t, 3, row + count, 0, mean, NULL); @@ -914,6 +1153,7 @@ show_descriptives (const struct oneway_spec *cmd, const struct oneway_workspace tab_double (t, 9, row + count, 0, dd->maximum, fmt); } + if (categoricals_is_complete (cats)) { double T; double n, mean, variance; @@ -945,12 +1185,13 @@ show_descriptives (const struct oneway_spec *cmd, const struct oneway_workspace tab_double (t, 7, row + count, 0, mean + T * std_error, NULL); + /* Min and Max */ tab_double (t, 8, row + count, 0, ws->dd_total[v]->minimum, fmt); tab_double (t, 9, row + count, 0, ws->dd_total[v]->maximum, fmt); } - row += categoricals_total (cats) + 1; + row += categoricals_n_total (cats) + 1; } tab_submit (t); @@ -987,19 +1228,20 @@ show_homogeneity (const struct oneway_spec *cmd, const struct oneway_workspace * for (v = 0; v < cmd->n_vars; ++v) { + double n; const struct per_var_ws *pvw = &ws->vws[v]; - const struct categoricals *cats = covariance_get_categoricals (pvw->cov); + double F = levene_calculate (pvw->nl); const struct variable *var = cmd->vars[v]; - const struct group_proc *gp = group_proc_get (cmd->vars[v]); const char *s = var_to_string (var); + double df1, df2; - const double df1 = pvw->n_groups - 1; - const double df2 = pvw->cc - pvw->n_groups; - double F = gp->levene; + moments1_calculate (ws->dd_total[v]->mom, &n, NULL, NULL, NULL, NULL); - tab_text (t, 0, v + 1, TAB_LEFT | TAT_TITLE, s); + df1 = pvw->n_groups - 1; + df2 = n - pvw->n_groups; + tab_text (t, 0, v + 1, TAB_LEFT | TAT_TITLE, s); tab_double (t, 1, v + 1, TAB_RIGHT, F, NULL); tab_fixed (t, 2, v + 1, TAB_RIGHT, df1, 8, 0); @@ -1078,25 +1320,19 @@ show_contrast_coeffs (const struct oneway_spec *cmd, const struct oneway_workspa ++count, coeffi = ll_next (coeffi)) { const struct categoricals *cats = covariance_get_categoricals (cov); - const union value *val = categoricals_get_value_by_subscript (cats, count); + const struct ccase *gcc = categoricals_get_case_by_category (cats, count); + struct coeff_node *coeffn = ll_data (coeffi, struct coeff_node, ll); struct string vstr; ds_init_empty (&vstr); - var_append_value_name (cmd->indep_var, val, &vstr); + var_append_value_name (cmd->indep_var, case_data (gcc, cmd->indep_var), &vstr); tab_text (t, count + 2, 1, TAB_CENTER | TAT_TITLE, ds_cstr (&vstr)); ds_destroy (&vstr); - if (cn->bad_count) - tab_text (t, count + 2, c_num + 2, TAB_RIGHT, "?" ); - else - { - struct coeff_node *coeffn = ll_data (coeffi, struct coeff_node, ll); - - tab_text_format (t, count + 2, c_num + 2, TAB_RIGHT, "%g", coeffn->coeff); - } + tab_text_format (t, count + 2, c_num + 2, TAB_RIGHT, "%g", coeffn->coeff); } ++c_num; } @@ -1207,22 +1443,19 @@ show_contrast_tests (const struct oneway_spec *cmd, const struct oneway_workspac (v * lines_per_variable) + i + 1 + n_contrasts, TAB_CENTER | TAT_TITLE, "%d", i + 1); - if (cn->bad_count) - continue; - for (coeffi = ll_head (&cn->coefficient_list); coeffi != ll_null (&cn->coefficient_list); ++ci, coeffi = ll_next (coeffi)) { double n, mean, variance; - const struct descriptive_data *dd = categoricals_get_user_data_by_subscript (cats, ci); - - moments1_calculate (dd->mom, &n, &mean, &variance, NULL, NULL); - + const struct descriptive_data *dd = categoricals_get_user_data_by_category (cats, ci); struct coeff_node *cn = ll_data (coeffi, struct coeff_node, ll); const double coef = cn->coeff; + double winv ; - const double winv = variance / n; + moments1_calculate (dd->mom, &n, &mean, &variance, NULL, NULL); + + winv = variance / n; contrast_value += coef * mean; @@ -1307,3 +1540,127 @@ show_contrast_tests (const struct oneway_spec *cmd, const struct oneway_workspac } + +static void +show_comparisons (const struct oneway_spec *cmd, const struct oneway_workspace *ws, int v) +{ + const int n_cols = 8; + const int heading_rows = 2; + const int heading_cols = 3; + + int p; + int r = heading_rows ; + + const struct per_var_ws *pvw = &ws->vws[v]; + const struct categoricals *cat = pvw->cat; + const int n_rows = heading_rows + cmd->n_posthoc * pvw->n_groups * (pvw->n_groups - 1); + + struct tab_table *t = tab_create (n_cols, n_rows); + + tab_headers (t, heading_cols, 0, heading_rows, 0); + + /* Put a frame around the entire box, and vertical lines inside */ + tab_box (t, + TAL_2, TAL_2, + -1, -1, + 0, 0, + n_cols - 1, n_rows - 1); + + tab_box (t, + -1, -1, + -1, TAL_1, + heading_cols, 0, + n_cols - 1, n_rows - 1); + + tab_vline (t, TAL_2, heading_cols, 0, n_rows - 1); + + tab_title (t, _("Multiple Comparisons (%s)"), var_to_string (cmd->vars[v])); + + tab_text_format (t, 1, 1, TAB_LEFT | TAT_TITLE, _("(I) %s"), var_to_string (cmd->indep_var)); + tab_text_format (t, 2, 1, TAB_LEFT | TAT_TITLE, _("(J) %s"), var_to_string (cmd->indep_var)); + tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Mean Difference")); + tab_text (t, 3, 1, TAB_CENTER | TAT_TITLE, _("(I - J)")); + tab_text (t, 4, 1, TAB_CENTER | TAT_TITLE, _("Std. Error")); + tab_text (t, 5, 1, TAB_CENTER | TAT_TITLE, _("Sig.")); + + tab_joint_text_format (t, 6, 0, 7, 0, TAB_CENTER | TAT_TITLE, + _("%g%% Confidence Interval"), + (1 - cmd->alpha) * 100.0); + + tab_text (t, 6, 1, TAB_CENTER | TAT_TITLE, _("Lower Bound")); + tab_text (t, 7, 1, TAB_CENTER | TAT_TITLE, _("Upper Bound")); + + + for (p = 0; p < cmd->n_posthoc; ++p) + { + int i; + const struct posthoc *ph = &ph_tests[cmd->posthoc[p]]; + + tab_hline (t, TAL_2, 0, n_cols - 1, r); + + tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, gettext (ph->label)); + + for (i = 0; i < pvw->n_groups ; ++i) + { + double weight_i, mean_i, var_i; + int rx = 0; + struct string vstr; + int j; + struct descriptive_data *dd_i = categoricals_get_user_data_by_category (cat, i); + const struct ccase *gcc = categoricals_get_case_by_category (cat, i); + + + ds_init_empty (&vstr); + var_append_value_name (cmd->indep_var, case_data (gcc, cmd->indep_var), &vstr); + + if ( i != 0) + tab_hline (t, TAL_1, 1, n_cols - 1, r); + tab_text (t, 1, r, TAB_LEFT | TAT_TITLE, ds_cstr (&vstr)); + + moments1_calculate (dd_i->mom, &weight_i, &mean_i, &var_i, 0, 0); + + for (j = 0 ; j < pvw->n_groups; ++j) + { + double std_err; + double weight_j, mean_j, var_j; + double half_range; + const struct ccase *cc; + struct descriptive_data *dd_j = categoricals_get_user_data_by_category (cat, j); + if (j == i) + continue; + + ds_clear (&vstr); + cc = categoricals_get_case_by_category (cat, j); + var_append_value_name (cmd->indep_var, case_data (cc, cmd->indep_var), &vstr); + tab_text (t, 2, r + rx, TAB_LEFT | TAT_TITLE, ds_cstr (&vstr)); + + moments1_calculate (dd_j->mom, &weight_j, &mean_j, &var_j, 0, 0); + + tab_double (t, 3, r + rx, 0, mean_i - mean_j, 0); + + std_err = pvw->mse; + std_err *= weight_i + weight_j; + std_err /= weight_i * weight_j; + std_err = sqrt (std_err); + + tab_double (t, 4, r + rx, 0, std_err, 0); + + tab_double (t, 5, r + rx, 0, 2 * multiple_comparison_sig (std_err, pvw, dd_i, dd_j, ph), 0); + + half_range = mc_half_range (cmd, pvw, std_err, dd_i, dd_j, ph); + + tab_double (t, 6, r + rx, 0, + (mean_i - mean_j) - half_range, 0 ); + + tab_double (t, 7, r + rx, 0, + (mean_i - mean_j) + half_range, 0 ); + + rx++; + } + ds_destroy (&vstr); + r += pvw->n_groups - 1; + } + } + + tab_submit (t); +}