/* 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, 2014,
+ 2020 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
#include <config.h>
-#include <data/case.h>
-#include <data/casegrouper.h>
-#include <data/casereader.h>
-
-#include <math/covariance.h>
-#include <math/categoricals.h>
+#include <float.h>
+#include <gsl/gsl_cdf.h>
#include <gsl/gsl_matrix.h>
-#include <linreg/sweep.h>
+#include <math.h>
-#include <libpspp/ll.h>
+#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/pivot-table.h"
-#include <language/lexer/lexer.h>
-#include <language/lexer/variable-parser.h>
-#include <language/lexer/value-parser.h>
-#include <language/command.h>
+#include "gettext.h"
+#define _(msgid) gettext (msgid)
+#define N_(msgid) msgid
-#include <data/procedure.h>
-#include <data/dictionary.h>
+/* Workspace variable for each dependent variable */
+struct per_var_ws
+{
+ struct interaction *iact;
+ struct categoricals *cat;
+ struct covariance *cov;
+ struct levene *nl;
+ double n;
-#include <language/dictionary/split-file.h>
-#include <libpspp/hash.h>
-#include <libpspp/taint.h>
-#include <math/group-proc.h>
-#include <math/levene.h>
-#include <libpspp/misc.h>
+ double sst;
+ double sse;
+ double ssa;
-#include <output/tab.h>
+ int n_groups;
-#include <gsl/gsl_cdf.h>
-#include <math.h>
-#include <data/format.h>
+ double mse;
+};
-#include <libpspp/message.h>
+/* Per category data */
+struct descriptive_data
+{
+ const struct variable *var;
+ struct moments1 *mom;
-#include "gettext.h"
-#define _(msgid) gettext (msgid)
+ double minimum;
+ double maximum;
+};
enum missing_type
{
struct coeff_node
{
- struct ll ll;
- double coeff;
+ struct ll ll;
+ double coeff;
};
struct contrasts_node
{
- struct ll ll;
+ 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
/* The weight variable */
const struct variable *wv;
+ const struct fmt_spec *wfmt;
+
+ /* 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);
- int n_groups;
+ if (n_i <= 1.0 || n_j <= 1.0)
+ return SYSMIS;
+
+ 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;
-};
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 */
+ struct descriptive_data **dd_total;
};
/* Routines to show the output tables */
static void show_anova_table (const struct oneway_spec *, const struct oneway_workspace *);
-static void show_descriptives (const struct oneway_spec *);
-static void show_homogeneity (const struct oneway_spec *);
+static void show_descriptives (const struct oneway_spec *, const struct oneway_workspace *);
+static void show_homogeneity (const struct oneway_spec *, const struct oneway_workspace *);
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)
{
- const struct dictionary *dict = dataset_dict (ds);
+ const struct dictionary *dict = dataset_dict (ds);
struct oneway_spec oneway ;
oneway.n_vars = 0;
oneway.vars = NULL;
oneway.missing_type = MISS_ANALYSIS;
oneway.exclude = MV_ANY;
oneway.wv = dict_get_weight (dict);
+ oneway.wfmt = dict_get_weight_format (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,
PV_NO_DUPLICATE | PV_NUMERIC))
goto error;
- lex_force_match (lexer, T_BY);
+ if (!lex_force_match (lexer, T_BY))
+ goto error;
oneway.indep_var = parse_variable_const (lexer, dict);
+ if (oneway.indep_var == NULL)
+ goto error;
- 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"))
{
}
}
}
+ 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;
+ if (! lex_force_num (lexer))
+ goto error;
+ 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 contrasts_node *cl = XZALLOC (struct contrasts_node);
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))
+ if (lex_is_number (lexer))
{
struct coeff_node *cc = xmalloc (sizeof *cc);
cc->coeff = lex_number (lexer);
}
else
{
+ destroy_coeff_list (cl);
lex_error (lexer, NULL);
goto error;
}
}
+ if (ll_count (coefficient_list) <= 0)
+ {
+ destroy_coeff_list (cl);
+ goto error;
+ }
+
ll_push_tail (&oneway.contrast_list, &cl->ll);
}
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"))
{
struct casereader *group;
bool ok;
-
-
grouper = casegrouper_create_splits (proc_open (ds), dict);
while (casegrouper_get_next_group (grouper, &group))
run_oneway (&oneway, group, 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;
}
\f
-static int
-compare_double_3way (const void *a_, const void *b_, const void *aux UNUSED)
+
+static struct descriptive_data *
+dd_create (const struct variable *var)
+{
+ struct descriptive_data *dd = xmalloc (sizeof *dd);
+
+ dd->mom = moments1_create (MOMENT_VARIANCE);
+ dd->minimum = DBL_MAX;
+ dd->maximum = -DBL_MAX;
+ dd->var = var;
+
+ return dd;
+}
+
+static void
+dd_destroy (struct descriptive_data *dd)
{
- const double *a = a_;
- const double *b = b_;
- return *a < *b ? -1 : *a > *b;
+ moments1_destroy (dd->mom);
+ free (dd);
}
-static unsigned
-do_hash_double (const void *value_, const void *aux UNUSED)
+static void *
+makeit (const void *aux1, void *aux2 UNUSED)
{
- const double *value = value_;
- return hash_double (*value, 0);
+ const struct variable *var = aux1;
+
+ struct descriptive_data *dd = dd_create (var);
+
+ return dd;
}
static void
-free_double (void *value_, const void *aux UNUSED)
+killit (const void *aux1 UNUSED, void *aux2 UNUSED, void *user_data)
{
- double *value = value_;
- free (value);
+ struct descriptive_data *dd = user_data;
+
+ dd_destroy (dd);
}
-static void postcalc (const struct oneway_spec *cmd);
-static void precalc (const struct oneway_spec *cmd);
+static void
+updateit (const void *aux1, void *aux2, void *user_data,
+ const struct ccase *c, double weight)
+{
+ struct descriptive_data *dd = user_data;
+
+ const struct variable *varp = aux1;
+
+ const union value *valx = case_data (c, varp);
+
+ struct descriptive_data *dd_total = aux2;
+
+ moments1_add (dd->mom, valx->f, weight);
+ if (valx->f < dd->minimum)
+ dd->minimum = valx->f;
+
+ if (valx->f > dd->maximum)
+ dd->maximum = valx->f;
+
+ {
+ const struct variable *var = dd_total->var;
+ const union value *val = case_data (c, var);
+
+ moments1_add (dd_total->mom,
+ val->f,
+ weight);
+
+ if (val->f < dd_total->minimum)
+ dd_total->minimum = val->f;
+
+ if (val->f > dd_total->maximum)
+ dd_total->maximum = val->f;
+ }
+}
static void
run_oneway (const struct oneway_spec *cmd,
struct casereader *reader;
struct ccase *c;
-
struct oneway_workspace ws;
- ws.vws = xmalloc (cmd->n_vars * sizeof (*ws.vws));
+ ws.actual_number_of_groups = 0;
+ ws.vws = xcalloc (cmd->n_vars, sizeof (*ws.vws));
+ ws.dd_total = XCALLOC (cmd->n_vars, struct descriptive_data*);
+ 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);
+ static const struct payload payload =
+ {
+ .create = makeit,
+ .update = updateit,
+ .calculate = NULL,
+ .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);
+
+ 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,
- cmd->wv, cmd->exclude);
- ws.vws[v].cc = 0;
+ ws.vws[v].cat,
+ cmd->wv, cmd->exclude, true);
+ 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)
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))
+ 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->n - pvw->n_groups);
}
- postcalc (cmd);
+ for (v = 0; v < cmd->n_vars; ++v)
+ {
+ const struct categoricals *cats = covariance_get_categoricals (ws.vws[v].cov);
- if ( cmd->stats & STATS_HOMOGENEITY )
- levene (dict, casereader_clone (input), cmd->indep_var,
- cmd->n_vars, cmd->vars, cmd->exclude);
+ if (! categoricals_is_complete (cats))
+ {
+ continue;
+ }
- casereader_destroy (input);
+ if (categoricals_n_total (cats) > ws.actual_number_of_groups)
+ ws.actual_number_of_groups = categoricals_n_total (cats);
+ }
- ws.actual_number_of_groups = hsh_count (ws.group_hash);
+ casereader_destroy (input);
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, struct oneway_workspace *ws);
-static void show_contrast_tests (const struct oneway_spec *cmd, struct oneway_workspace *ws);
+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)
/* 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;
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;
}
ll_for_each (cn, struct coeff_node, ll, cl)
sum += cn->coeff;
- if ( sum != 0.0 )
+ if (sum != 0.0)
msg (SW, _("Coefficients for contrast %zu do not total zero"), i);
}
if (cmd->stats & STATS_DESCRIPTIVES)
- show_descriptives (cmd);
+ show_descriptives (cmd, ws);
if (cmd->stats & STATS_HOMOGENEITY)
- show_homogeneity (cmd);
+ show_homogeneity (cmd, 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);
}
static void
show_anova_table (const struct oneway_spec *cmd, const struct oneway_workspace *ws)
{
- size_t i;
- int n_cols =7;
- size_t n_rows = cmd->n_vars * 3 + 1;
-
- struct tab_table *t = tab_create (n_cols, n_rows);
-
- tab_headers (t, 2, 0, 1, 0);
-
- tab_box (t,
- TAL_2, TAL_2,
- -1, TAL_1,
- 0, 0,
- n_cols - 1, n_rows - 1);
+ struct pivot_table *table = pivot_table_create (N_("ANOVA"));
- tab_hline (t, TAL_2, 0, n_cols - 1, 1 );
- tab_vline (t, TAL_2, 2, 0, n_rows - 1);
- tab_vline (t, TAL_0, 1, 0, 0);
+ pivot_dimension_create (table, PIVOT_AXIS_COLUMN, N_("Statistics"),
+ N_("Sum of Squares"), PIVOT_RC_OTHER,
+ N_("df"), PIVOT_RC_INTEGER,
+ N_("Mean Square"), PIVOT_RC_OTHER,
+ N_("F"), PIVOT_RC_OTHER,
+ N_("Sig."), PIVOT_RC_SIGNIFICANCE);
- tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("Sum of Squares"));
- tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("df"));
- tab_text (t, 4, 0, TAB_CENTER | TAT_TITLE, _("Mean Square"));
- tab_text (t, 5, 0, TAB_CENTER | TAT_TITLE, _("F"));
- tab_text (t, 6, 0, TAB_CENTER | TAT_TITLE, _("Significance"));
+ pivot_dimension_create (table, PIVOT_AXIS_ROW, N_("Type"),
+ N_("Between Groups"), N_("Within Groups"),
+ N_("Total"));
+ struct pivot_dimension *variables = pivot_dimension_create (
+ table, PIVOT_AXIS_ROW, N_("Variables"));
- for (i = 0; i < cmd->n_vars; ++i)
+ for (size_t i = 0; i < cmd->n_vars; ++i)
{
- const struct per_var_ws *pvw = &ws->vws[i];
- struct group_proc *gp = group_proc_get (cmd->vars[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]);
-
- 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"));
- tab_text (t, 1, i * 3 + 2, TAB_LEFT | TAT_TITLE, _("Within Groups"));
- tab_text (t, 1, i * 3 + 3, TAB_LEFT | TAT_TITLE, _("Total"));
+ int var_idx = pivot_category_create_leaf (
+ variables->root, pivot_value_new_variable (cmd->vars[i]));
- if (i > 0)
- tab_hline (t, TAL_1, 0, n_cols - 1, i * 3 + 1);
-
-
- gp->mse = (pvw->sst - pvw->ssa) / df2;
-
- /* Sums of Squares */
- tab_double (t, 2, i * 3 + 1, 0, pvw->ssa, NULL);
- tab_double (t, 2, i * 3 + 3, 0, pvw->sst, NULL);
- tab_double (t, 2, i * 3 + 2, 0, pvw->sse, NULL);
-
-
- /* 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);
-
- /* Mean Squares */
- tab_double (t, 4, i * 3 + 1, TAB_RIGHT, msa, NULL);
- tab_double (t, 4, i * 3 + 2, TAB_RIGHT, gp->mse, NULL);
+ const struct per_var_ws *pvw = &ws->vws[i];
- {
- const double F = msa / gp->mse ;
+ double n;
+ moments1_calculate (ws->dd_total[i]->mom, &n, NULL, NULL, NULL, NULL);
- /* The F value */
- tab_double (t, 5, i * 3 + 1, 0, F, NULL);
+ double df1 = pvw->n_groups - 1;
+ double df2 = n - pvw->n_groups;
+ double msa = pvw->ssa / df1;
+ double F = msa / pvw->mse ;
- /* The significance */
- tab_double (t, 6, i * 3 + 1, 0, gsl_cdf_fdist_Q (F, df1, df2), NULL);
- }
+ struct entry
+ {
+ int stat_idx;
+ int type_idx;
+ double x;
+ }
+ entries[] = {
+ /* Sums of Squares. */
+ { 0, 0, pvw->ssa },
+ { 0, 1, pvw->sse },
+ { 0, 2, pvw->sst },
+ /* Degrees of Freedom. */
+ { 1, 0, df1 },
+ { 1, 1, df2 },
+ { 1, 2, n - 1 },
+ /* Mean Squares. */
+ { 2, 0, msa },
+ { 2, 1, pvw->mse },
+ /* F. */
+ { 3, 0, F },
+ /* Significance. */
+ { 4, 0, gsl_cdf_fdist_Q (F, df1, df2) },
+ };
+ for (size_t j = 0; j < sizeof entries / sizeof *entries; j++)
+ {
+ const struct entry *e = &entries[j];
+ pivot_table_put3 (table, e->stat_idx, e->type_idx, var_idx,
+ pivot_value_new_number (e->x));
+ }
}
- tab_title (t, _("ANOVA"));
- tab_submit (t);
+ pivot_table_submit (table);
}
-
/* Show the descriptives table */
static void
-show_descriptives (const struct oneway_spec *cmd)
+show_descriptives (const struct oneway_spec *cmd, const struct oneway_workspace *ws)
{
- size_t v;
- int n_cols = 10;
- struct tab_table *t;
- int row;
-
- const double confidence = 0.95;
- const double q = (1.0 - confidence) / 2.0;
-
- const struct fmt_spec *wfmt = cmd->wv ? var_get_print_format (cmd->wv) : & F_8_0;
-
- int n_rows = 2;
-
- for ( v = 0; v < cmd->n_vars; ++v )
- n_rows += group_proc_get (cmd->vars[v])->n_groups + 1;
-
- t = tab_create (n_cols, n_rows);
- tab_headers (t, 2, 0, 2, 0);
-
-
- /* Put a frame around the entire box, and vertical lines inside */
- tab_box (t,
- TAL_2, TAL_2,
- -1, TAL_1,
- 0, 0,
- n_cols - 1, n_rows - 1);
-
- /* Underline headers */
- tab_hline (t, TAL_2, 0, n_cols - 1, 2);
- tab_vline (t, TAL_2, 2, 0, n_rows - 1);
-
- tab_text (t, 2, 1, TAB_CENTER | TAT_TITLE, _("N"));
- tab_text (t, 3, 1, TAB_CENTER | TAT_TITLE, _("Mean"));
- tab_text (t, 4, 1, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
- tab_text (t, 5, 1, TAB_CENTER | TAT_TITLE, _("Std. Error"));
-
-
- tab_vline (t, TAL_0, 7, 0, 0);
- tab_hline (t, TAL_1, 6, 7, 1);
- tab_joint_text_format (t, 6, 0, 7, 0, TAB_CENTER | TAT_TITLE,
- _("%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"));
-
- tab_text (t, 8, 1, TAB_CENTER | TAT_TITLE, _("Minimum"));
- tab_text (t, 9, 1, TAB_CENTER | TAT_TITLE, _("Maximum"));
+ if (!cmd->n_vars)
+ return;
+ const struct categoricals *cats = covariance_get_categoricals (
+ ws->vws[0].cov);
+ struct pivot_table *table = pivot_table_create (N_("Descriptives"));
+ pivot_table_set_weight_format (table, cmd->wfmt);
- tab_title (t, _("Descriptives"));
+ const double confidence = 0.95;
+ struct pivot_dimension *statistics = pivot_dimension_create (
+ table, PIVOT_AXIS_COLUMN, N_("Statistics"),
+ N_("N"), PIVOT_RC_COUNT, N_("Mean"), N_("Std. Deviation"),
+ N_("Std. Error"));
+ struct pivot_category *interval = pivot_category_create_group__ (
+ statistics->root,
+ pivot_value_new_text_format (N_("%g%% Confidence Interval for Mean"),
+ confidence * 100.0));
+ pivot_category_create_leaves (interval, N_("Lower Bound"),
+ N_("Upper Bound"));
+ pivot_category_create_leaves (statistics->root,
+ N_("Minimum"), N_("Maximum"));
+
+ struct pivot_dimension *indep_var = pivot_dimension_create__ (
+ table, PIVOT_AXIS_ROW, pivot_value_new_variable (cmd->indep_var));
+ indep_var->root->show_label = true;
+ size_t n;
+ union value *values = categoricals_get_var_values (cats, cmd->indep_var, &n);
+ for (size_t j = 0; j < n; j++)
+ pivot_category_create_leaf (
+ indep_var->root, pivot_value_new_var_value (cmd->indep_var, &values[j]));
+ pivot_category_create_leaf (
+ indep_var->root, pivot_value_new_text_format (N_("Total")));
+
+ struct pivot_dimension *dep_var = pivot_dimension_create (
+ table, PIVOT_AXIS_ROW, N_("Dependent Variable"));
- row = 2;
- for (v = 0; v < cmd->n_vars; ++v)
+ const double q = (1.0 - confidence) / 2.0;
+ for (int v = 0; v < cmd->n_vars; ++v)
{
- double T;
- double std_error;
-
- struct group_proc *gp = group_proc_get (cmd->vars[v]);
-
- struct group_statistics *gs;
- struct group_statistics *totals = &gp->ugs;
-
- const char *s = var_to_string (cmd->vars[v]);
- const struct fmt_spec *fmt = var_get_print_format (cmd->vars[v]);
-
- struct group_statistics *const *gs_array =
- (struct group_statistics *const *) hsh_sort (gp->group_hash);
- int count = 0;
+ int dep_var_idx = pivot_category_create_leaf (
+ dep_var->root, pivot_value_new_variable (cmd->vars[v]));
- tab_text (t, 0, row, TAB_LEFT | TAT_TITLE, s);
- if ( v > 0)
- tab_hline (t, TAL_1, 0, n_cols - 1, row);
+ struct per_var_ws *pvw = &ws->vws[v];
+ const struct categoricals *cats = covariance_get_categoricals (pvw->cov);
- for (count = 0; count < hsh_count (gp->group_hash); ++count)
+ int count;
+ for (count = 0; count < categoricals_n_total (cats); ++count)
{
- struct string vstr;
- ds_init_empty (&vstr);
- gs = gs_array[count];
-
- var_append_value_name (cmd->indep_var, &gs->id, &vstr);
-
- tab_text (t, 1, row + count,
- TAB_LEFT | TAT_TITLE,
- ds_cstr (&vstr));
-
- ds_destroy (&vstr);
-
- /* Now fill in the numbers ... */
-
- tab_fixed (t, 2, row + count, 0, gs->n, 8, 0);
-
- tab_double (t, 3, row + count, 0, gs->mean, NULL);
-
- tab_double (t, 4, row + count, 0, gs->std_dev, NULL);
-
- std_error = gs->std_dev / sqrt (gs->n) ;
- tab_double (t, 5, row + count, 0,
- std_error, NULL);
-
- /* Now the confidence interval */
-
- T = gsl_cdf_tdist_Qinv (q, gs->n - 1);
-
- tab_double (t, 6, row + count, 0,
- gs->mean - T * std_error, NULL);
-
- tab_double (t, 7, row + count, 0,
- gs->mean + T * std_error, NULL);
-
- /* Min and Max */
-
- tab_double (t, 8, row + count, 0, gs->minimum, fmt);
- tab_double (t, 9, row + count, 0, gs->maximum, fmt);
+ const struct descriptive_data *dd
+ = categoricals_get_user_data_by_category (cats, count);
+
+ double n, mean, variance;
+ moments1_calculate (dd->mom, &n, &mean, &variance, NULL, NULL);
+
+ double std_dev = sqrt (variance);
+ double std_error = std_dev / sqrt (n) ;
+ double T = gsl_cdf_tdist_Qinv (q, n - 1);
+
+ double entries[] = {
+ n,
+ mean,
+ std_dev,
+ std_error,
+ mean - T * std_error,
+ mean + T * std_error,
+ dd->minimum,
+ dd->maximum,
+ };
+ for (size_t i = 0; i < sizeof entries / sizeof *entries; i++)
+ pivot_table_put3 (table, i, count, dep_var_idx,
+ pivot_value_new_number (entries[i]));
}
- tab_text (t, 1, row + count,
- TAB_LEFT | TAT_TITLE, _("Total"));
-
- tab_double (t, 2, row + count, 0, totals->n, wfmt);
-
- tab_double (t, 3, row + count, 0, totals->mean, NULL);
-
- tab_double (t, 4, row + count, 0, totals->std_dev, NULL);
-
- std_error = totals->std_dev / sqrt (totals->n) ;
-
- tab_double (t, 5, row + count, 0, std_error, NULL);
-
- /* Now the confidence interval */
-
- T = gsl_cdf_tdist_Qinv (q, totals->n - 1);
-
- tab_double (t, 6, row + count, 0,
- totals->mean - T * std_error, NULL);
-
- tab_double (t, 7, row + count, 0,
- totals->mean + T * std_error, NULL);
-
- /* Min and Max */
-
- tab_double (t, 8, row + count, 0, totals->minimum, fmt);
- tab_double (t, 9, row + count, 0, totals->maximum, fmt);
-
- row += gp->n_groups + 1;
+ if (categoricals_is_complete (cats))
+ {
+ double n, mean, variance;
+ moments1_calculate (ws->dd_total[v]->mom, &n, &mean, &variance,
+ NULL, NULL);
+
+ double std_dev = sqrt (variance);
+ double std_error = std_dev / sqrt (n) ;
+ double T = gsl_cdf_tdist_Qinv (q, n - 1);
+
+ double entries[] = {
+ n,
+ mean,
+ std_dev,
+ std_error,
+ mean - T * std_error,
+ mean + T * std_error,
+ ws->dd_total[v]->minimum,
+ ws->dd_total[v]->maximum,
+ };
+ for (size_t i = 0; i < sizeof entries / sizeof *entries; i++)
+ pivot_table_put3 (table, i, count, dep_var_idx,
+ pivot_value_new_number (entries[i]));
+ }
}
- tab_submit (t);
+ pivot_table_submit (table);
}
/* Show the homogeneity table */
static void
-show_homogeneity (const struct oneway_spec *cmd)
+show_homogeneity (const struct oneway_spec *cmd, const struct oneway_workspace *ws)
{
- size_t v;
- int n_cols = 5;
- size_t n_rows = cmd->n_vars + 1;
-
- struct tab_table *t;
-
-
- t = tab_create (n_cols, n_rows);
- tab_headers (t, 1, 0, 1, 0);
-
+ struct pivot_table *table = pivot_table_create (
+ N_("Test of Homogeneity of Variances"));
- /* Put a frame around the entire box, and vertical lines inside */
- tab_box (t,
- TAL_2, TAL_2,
- -1, TAL_1,
- 0, 0,
- n_cols - 1, n_rows - 1);
+ pivot_dimension_create (table, PIVOT_AXIS_COLUMN, N_("Statistics"),
+ N_("Levene Statistic"), PIVOT_RC_OTHER,
+ N_("df1"), PIVOT_RC_INTEGER,
+ N_("df2"), PIVOT_RC_INTEGER,
+ N_("Sig."), PIVOT_RC_SIGNIFICANCE);
+ struct pivot_dimension *variables = pivot_dimension_create (
+ table, PIVOT_AXIS_ROW, N_("Variables"));
- tab_hline (t, TAL_2, 0, n_cols - 1, 1);
- tab_vline (t, TAL_2, 1, 0, n_rows - 1);
-
-
- tab_text (t, 1, 0, TAB_CENTER | TAT_TITLE, _("Levene Statistic"));
- tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("df1"));
- tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("df2"));
- tab_text (t, 4, 0, TAB_CENTER | TAT_TITLE, _("Significance"));
-
- tab_title (t, _("Test of Homogeneity of Variances"));
-
- for (v = 0; v < cmd->n_vars; ++v)
+ for (int v = 0; v < cmd->n_vars; ++v)
{
- double F;
- 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);
- const struct group_statistics *totals = &gp->ugs;
-
- const double df1 = gp->n_groups - 1;
- const double df2 = totals->n - gp->n_groups;
+ int var_idx = pivot_category_create_leaf (
+ variables->root, pivot_value_new_variable (cmd->vars[v]));
- tab_text (t, 0, v + 1, TAB_LEFT | TAT_TITLE, s);
+ double n;
+ moments1_calculate (ws->dd_total[v]->mom, &n, NULL, NULL, NULL, NULL);
- F = gp->levene;
- tab_double (t, 1, v + 1, TAB_RIGHT, F, NULL);
- tab_fixed (t, 2, v + 1, TAB_RIGHT, df1, 8, 0);
- tab_fixed (t, 3, v + 1, TAB_RIGHT, df2, 8, 0);
+ const struct per_var_ws *pvw = &ws->vws[v];
+ double df1 = pvw->n_groups - 1;
+ double df2 = n - pvw->n_groups;
+ double F = levene_calculate (pvw->nl);
- /* Now the significance */
- tab_double (t, 4, v + 1, TAB_RIGHT,gsl_cdf_fdist_Q (F, df1, df2), NULL);
+ double entries[] =
+ {
+ F,
+ df1,
+ df2,
+ gsl_cdf_fdist_Q (F, df1, df2),
+ };
+ for (size_t i = 0; i < sizeof entries / sizeof *entries; i++)
+ pivot_table_put2 (table, i, var_idx,
+ pivot_value_new_number (entries[i]));
}
- tab_submit (t);
+ pivot_table_submit (table);
}
/* Show the contrast coefficients table */
static void
-show_contrast_coeffs (const struct oneway_spec *cmd, struct oneway_workspace *ws)
+show_contrast_coeffs (const struct oneway_spec *cmd, const struct oneway_workspace *ws)
{
- int c_num = 0;
- struct ll *cli;
-
- int n_contrasts = ll_count (&cmd->contrast_list);
- int n_cols = 2 + ws->actual_number_of_groups;
- int n_rows = 2 + n_contrasts;
-
- struct tab_table *t;
-
- const struct covariance *cov = ws->vws[0].cov ;
-
- t = tab_create (n_cols, n_rows);
- tab_headers (t, 2, 0, 2, 0);
-
- /* Put a frame around the entire box, and vertical lines inside */
- tab_box (t,
- TAL_2, TAL_2,
- -1, TAL_1,
- 0, 0,
- n_cols - 1, n_rows - 1);
-
- tab_box (t,
- -1, -1,
- TAL_0, TAL_0,
- 2, 0,
- n_cols - 1, 0);
+ struct pivot_table *table = pivot_table_create (N_("Contrast Coefficients"));
- tab_box (t,
- -1, -1,
- TAL_0, TAL_0,
- 0, 0,
- 1, 1);
+ struct pivot_dimension *indep_var = pivot_dimension_create__ (
+ table, PIVOT_AXIS_COLUMN, pivot_value_new_variable (cmd->indep_var));
+ indep_var->root->show_label = true;
- tab_hline (t, TAL_1, 2, n_cols - 1, 1);
- tab_hline (t, TAL_2, 0, n_cols - 1, 2);
+ struct pivot_dimension *contrast = pivot_dimension_create (
+ table, PIVOT_AXIS_ROW, N_("Contrast"));
+ contrast->root->show_label = true;
- tab_vline (t, TAL_2, 2, 0, n_rows - 1);
+ const struct covariance *cov = ws->vws[0].cov;
- tab_title (t, _("Contrast Coefficients"));
-
- tab_text (t, 0, 2, TAB_LEFT | TAT_TITLE, _("Contrast"));
-
-
- tab_joint_text (t, 2, 0, n_cols - 1, 0, TAB_CENTER | TAT_TITLE,
- var_to_string (cmd->indep_var));
-
- for ( cli = ll_head (&cmd->contrast_list);
- cli != ll_null (&cmd->contrast_list);
- cli = ll_next (cli))
+ const struct contrasts_node *cn;
+ int c_num = 1;
+ ll_for_each (cn, struct contrasts_node, ll, &cmd->contrast_list)
{
- int count = 0;
- struct contrasts_node *cn = ll_data (cli, struct contrasts_node, ll);
- struct ll *coeffi ;
+ int contrast_idx = pivot_category_create_leaf (
+ contrast->root, pivot_value_new_integer (c_num++));
- tab_text_format (t, 1, c_num + 2, TAB_CENTER, "%d", c_num + 1);
-
- for (coeffi = ll_head (&cn->coefficient_list);
- coeffi != ll_null (&cn->coefficient_list);
- ++count, coeffi = ll_next (coeffi))
+ const struct coeff_node *coeffn;
+ int indep_idx = 0;
+ ll_for_each (coeffn, struct coeff_node, ll, &cn->coefficient_list)
{
const struct categoricals *cats = covariance_get_categoricals (cov);
- const union value *val = categoricals_get_value_by_subscript (cats, count);
- struct string vstr;
-
- ds_init_empty (&vstr);
-
- var_append_value_name (cmd->indep_var, val, &vstr);
+ const struct ccase *gcc = categoricals_get_case_by_category (
+ cats, indep_idx);
- tab_text (t, count + 2, 1, TAB_CENTER | TAT_TITLE, ds_cstr (&vstr));
+ if (!contrast_idx)
+ pivot_category_create_leaf (
+ indep_var->root, pivot_value_new_var_value (
+ cmd->indep_var, case_data (gcc, cmd->indep_var)));
- 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);
- }
+ pivot_table_put2 (table, indep_idx++, contrast_idx,
+ pivot_value_new_integer (coeffn->coeff));
}
- ++c_num;
}
- tab_submit (t);
+ pivot_table_submit (table);
}
-
/* Show the results of the contrast tests */
static void
-show_contrast_tests (const struct oneway_spec *cmd, struct oneway_workspace *ws)
+show_contrast_tests (const struct oneway_spec *cmd, const struct oneway_workspace *ws)
{
+ struct pivot_table *table = pivot_table_create (N_("Contrast Tests"));
+
+ pivot_dimension_create (table, PIVOT_AXIS_COLUMN, N_("Statistics"),
+ N_("Value of Contrast"), PIVOT_RC_OTHER,
+ N_("Std. Error"), PIVOT_RC_OTHER,
+ N_("t"), PIVOT_RC_OTHER,
+ N_("df"), PIVOT_RC_OTHER,
+ N_("Sig. (2-tailed)"), PIVOT_RC_SIGNIFICANCE);
+
+ struct pivot_dimension *contrasts = pivot_dimension_create (
+ table, PIVOT_AXIS_ROW, N_("Contrast"));
+ contrasts->root->show_label = true;
int n_contrasts = ll_count (&cmd->contrast_list);
- size_t v;
- int n_cols = 8;
- size_t n_rows = 1 + cmd->n_vars * 2 * n_contrasts;
-
- struct tab_table *t;
+ for (int i = 1; i <= n_contrasts; i++)
+ pivot_category_create_leaf (contrasts->root, pivot_value_new_integer (i));
- t = tab_create (n_cols, n_rows);
- tab_headers (t, 3, 0, 1, 0);
+ pivot_dimension_create (table, PIVOT_AXIS_ROW, N_("Assumption"),
+ N_("Assume equal variances"),
+ N_("Does not assume equal variances"));
- /* Put a frame around the entire box, and vertical lines inside */
- tab_box (t,
- TAL_2, TAL_2,
- -1, TAL_1,
- 0, 0,
- n_cols - 1, n_rows - 1);
+ struct pivot_dimension *variables = pivot_dimension_create (
+ table, PIVOT_AXIS_ROW, N_("Dependent Variable"));
- tab_box (t,
- -1, -1,
- TAL_0, TAL_0,
- 0, 0,
- 2, 0);
-
- tab_hline (t, TAL_2, 0, n_cols - 1, 1);
- tab_vline (t, TAL_2, 3, 0, n_rows - 1);
-
-
- tab_title (t, _("Contrast Tests"));
-
- tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("Contrast"));
- tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Value of Contrast"));
- tab_text (t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Error"));
- tab_text (t, 5, 0, TAB_CENTER | TAT_TITLE, _("t"));
- tab_text (t, 6, 0, TAB_CENTER | TAT_TITLE, _("df"));
- tab_text (t, 7, 0, TAB_CENTER | TAT_TITLE, _("Sig. (2-tailed)"));
-
- for (v = 0; v < cmd->n_vars; ++v)
+ for (int v = 0; v < cmd->n_vars; ++v)
{
- struct ll *cli;
- int i = 0;
- int lines_per_variable = 2 * n_contrasts;
+ const struct per_var_ws *pvw = &ws->vws[v];
+ const struct categoricals *cats = covariance_get_categoricals (pvw->cov);
+ if (!categoricals_is_complete (cats))
+ continue;
- tab_text (t, 0, (v * lines_per_variable) + 1, TAB_LEFT | TAT_TITLE,
- var_to_string (cmd->vars[v]));
+ int var_idx = pivot_category_create_leaf (
+ variables->root, pivot_value_new_variable (cmd->vars[v]));
- for ( cli = ll_head (&cmd->contrast_list);
- cli != ll_null (&cmd->contrast_list);
- ++i, cli = ll_next (cli))
+ struct contrasts_node *cn;
+ int contrast_idx = 0;
+ ll_for_each (cn, struct contrasts_node, ll, &cmd->contrast_list)
{
- struct contrasts_node *cn = ll_data (cli, struct contrasts_node, ll);
- struct ll *coeffi = ll_head (&cn->coefficient_list);
- int ci;
- double contrast_value = 0.0;
- double coef_msq = 0.0;
- struct group_proc *grp_data = group_proc_get (cmd->vars[v]);
- struct hsh_table *group_hash = grp_data->group_hash;
-
- void *const *group_stat_array;
-
- 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!!
}
*/
+ double grand_n;
+ moments1_calculate (ws->dd_total[v]->mom, &grand_n, NULL, NULL,
+ NULL, NULL);
+ double df = grand_n - pvw->n_groups;
+
+ double contrast_value = 0.0;
+ double coef_msq = 0.0;
+ double sec_vneq = 0.0;
double df_denominator = 0.0;
double df_numerator = 0.0;
- if ( i == 0 )
+ struct coeff_node *coeffn;
+ int ci = 0;
+ ll_for_each (coeffn, struct coeff_node, ll, &cn->coefficient_list)
{
- tab_text (t, 1, (v * lines_per_variable) + i + 1,
- TAB_LEFT | TAT_TITLE,
- _("Assume equal variances"));
-
- tab_text (t, 1, (v * lines_per_variable) + i + 1 + n_contrasts,
- TAB_LEFT | TAT_TITLE,
- _("Does not assume equal"));
- }
-
- tab_text_format (t, 2, (v * lines_per_variable) + i + 1,
- TAB_CENTER | TAT_TITLE, "%d", i + 1);
-
-
- tab_text_format (t, 2,
- (v * lines_per_variable) + i + 1 + n_contrasts,
- TAB_CENTER | TAT_TITLE, "%d", i + 1);
-
- if (cn->bad_count)
- continue;
-
- group_stat_array = hsh_sort (group_hash);
-
- for (ci = 0;
- coeffi != ll_null (&cn->coefficient_list) &&
- ci < hsh_count (group_hash);
- ++ci, coeffi = ll_next (coeffi))
- {
- struct coeff_node *cn = ll_data (coeffi, struct coeff_node, ll);
- const double coef = cn->coeff;
- struct group_statistics *gs = group_stat_array[ci];
-
- const double winv = pow2 (gs->std_dev) / gs->n;
-
- contrast_value += coef * gs->mean;
+ const struct descriptive_data *dd
+ = categoricals_get_user_data_by_category (cats, ci);
+ const double coef = coeffn->coeff;
- coef_msq += (coef * coef) / gs->n;
+ double n, mean, variance;
+ moments1_calculate (dd->mom, &n, &mean, &variance, NULL, NULL);
- sec_vneq += (coef * coef) * pow2 (gs->std_dev) /gs->n;
+ double winv = variance / n;
+ contrast_value += coef * mean;
+ coef_msq += pow2 (coef) / n;
+ sec_vneq += pow2 (coef) * variance / n;
+ df_numerator += pow2 (coef) * winv;
+ df_denominator += pow2(pow2 (coef) * winv) / (n - 1);
- df_numerator += (coef * coef) * winv;
- df_denominator += pow2((coef * coef) * winv) / (gs->n - 1);
+ ci++;
}
-
sec_vneq = sqrt (sec_vneq);
-
df_numerator = pow2 (df_numerator);
- tab_double (t, 3, (v * lines_per_variable) + i + 1,
- TAB_RIGHT, contrast_value, NULL);
-
- tab_double (t, 3, (v * lines_per_variable) + i + 1 +
- n_contrasts,
- TAB_RIGHT, contrast_value, NULL);
-
- std_error_contrast = sqrt (grp_data->mse * coef_msq);
-
- /* Std. Error */
- tab_double (t, 4, (v * lines_per_variable) + i + 1,
- TAB_RIGHT, std_error_contrast,
- NULL);
-
- T = fabs (contrast_value / std_error_contrast);
-
- /* T Statistic */
+ double std_error_contrast = sqrt (pvw->mse * coef_msq);
+ double T = contrast_value / std_error_contrast;
+ double T_ne = contrast_value / sec_vneq;
+ double df_ne = df_numerator / df_denominator;
- tab_double (t, 5, (v * lines_per_variable) + i + 1,
- TAB_RIGHT, T,
- NULL);
-
- df = grp_data->ugs.n - grp_data->n_groups;
-
- /* Degrees of Freedom */
- tab_fixed (t, 6, (v * lines_per_variable) + i + 1,
- TAB_RIGHT, df,
- 8, 0);
+ struct entry
+ {
+ int stat_idx;
+ int assumption_idx;
+ double x;
+ }
+ entries[] =
+ {
+ /* Assume equal. */
+ { 0, 0, contrast_value },
+ { 1, 0, std_error_contrast },
+ { 2, 0, T },
+ { 3, 0, df },
+ { 4, 0, 2 * gsl_cdf_tdist_Q (fabs(T), df) },
+ /* Do not assume equal. */
+ { 0, 1, contrast_value },
+ { 1, 1, sec_vneq },
+ { 2, 1, T_ne },
+ { 3, 1, df_ne },
+ { 4, 1, 2 * gsl_cdf_tdist_Q (fabs(T_ne), df_ne) },
+ };
+
+ for (size_t i = 0; i < sizeof entries / sizeof *entries; i++)
+ {
+ const struct entry *e = &entries[i];
+ pivot_table_put4 (
+ table, e->stat_idx, contrast_idx, e->assumption_idx, var_idx,
+ pivot_value_new_number (e->x));
+ }
+ contrast_idx++;
+ }
+ }
- /* Significance TWO TAILED !!*/
- tab_double (t, 7, (v * lines_per_variable) + i + 1,
- TAB_RIGHT, 2 * gsl_cdf_tdist_Q (T, df),
- NULL);
+ pivot_table_submit (table);
+}
- /* Now for the Variances NOT Equal case */
+static void
+show_comparisons (const struct oneway_spec *cmd, const struct oneway_workspace *ws, int v)
+{
+ struct pivot_table *table = pivot_table_create__ (
+ pivot_value_new_user_text_nocopy (xasprintf (
+ _("Multiple Comparisons (%s)"),
+ var_to_string (cmd->vars[v]))),
+ "Multiple Comparisons");
+
+ struct pivot_dimension *statistics = pivot_dimension_create (
+ table, PIVOT_AXIS_COLUMN, N_("Statistics"),
+ N_("Mean Difference (I - J)"), PIVOT_RC_OTHER,
+ N_("Std. Error"), PIVOT_RC_OTHER,
+ N_("Sig."), PIVOT_RC_SIGNIFICANCE);
+ struct pivot_category *interval = pivot_category_create_group__ (
+ statistics->root,
+ pivot_value_new_text_format (N_("%g%% Confidence Interval"),
+ (1 - cmd->alpha) * 100.0));
+ pivot_category_create_leaves (interval,
+ N_("Lower Bound"), PIVOT_RC_OTHER,
+ N_("Upper Bound"), PIVOT_RC_OTHER);
+
+ struct pivot_dimension *j_family = pivot_dimension_create (
+ table, PIVOT_AXIS_ROW, N_("(J) Family"));
+ j_family->root->show_label = true;
+
+ struct pivot_dimension *i_family = pivot_dimension_create (
+ table, PIVOT_AXIS_ROW, N_("(J) Family"));
+ i_family->root->show_label = true;
+
+ const struct per_var_ws *pvw = &ws->vws[v];
+ const struct categoricals *cat = pvw->cat;
+ for (int i = 0; i < pvw->n_groups; i++)
+ {
+ const struct ccase *gcc = categoricals_get_case_by_category (cat, i);
+ for (int j = 0; j < 2; j++)
+ pivot_category_create_leaf (
+ j ? j_family->root : i_family->root,
+ pivot_value_new_var_value (cmd->indep_var,
+ case_data (gcc, cmd->indep_var)));
+ }
- /* Std. Error */
- tab_double (t, 4,
- (v * lines_per_variable) + i + 1 + n_contrasts,
- TAB_RIGHT, sec_vneq,
- NULL);
+ struct pivot_dimension *test = pivot_dimension_create (
+ table, PIVOT_AXIS_ROW, N_("Test"));
- T = contrast_value / sec_vneq;
- tab_double (t, 5,
- (v * lines_per_variable) + i + 1 + n_contrasts,
- TAB_RIGHT, T,
- NULL);
+ for (int p = 0; p < cmd->n_posthoc; ++p)
+ {
+ const struct posthoc *ph = &ph_tests[cmd->posthoc[p]];
- df = df_numerator / df_denominator;
+ int test_idx = pivot_category_create_leaf (
+ test->root, pivot_value_new_text (ph->label));
- tab_double (t, 6,
- (v * lines_per_variable) + i + 1 + n_contrasts,
- TAB_RIGHT, df,
- NULL);
+ for (int i = 0; i < pvw->n_groups ; ++i)
+ {
+ struct descriptive_data *dd_i
+ = categoricals_get_user_data_by_category (cat, i);
+ double weight_i, mean_i, var_i;
+ moments1_calculate (dd_i->mom, &weight_i, &mean_i, &var_i, 0, 0);
- /* The Significance */
- tab_double (t, 7, (v * lines_per_variable) + i + 1 + n_contrasts,
- TAB_RIGHT, 2 * gsl_cdf_tdist_Q (T,df),
- NULL);
+ for (int j = 0 ; j < pvw->n_groups; ++j)
+ {
+ if (j == i)
+ continue;
+
+ struct descriptive_data *dd_j
+ = categoricals_get_user_data_by_category (cat, j);
+ double weight_j, mean_j, var_j;
+ moments1_calculate (dd_j->mom, &weight_j, &mean_j, &var_j, 0, 0);
+
+ double std_err = pvw->mse;
+ std_err *= weight_i + weight_j;
+ std_err /= weight_i * weight_j;
+ std_err = sqrt (std_err);
+
+ double sig = 2 * multiple_comparison_sig (std_err, pvw,
+ dd_i, dd_j, ph);
+ double half_range = mc_half_range (cmd, pvw, std_err,
+ dd_i, dd_j, ph);
+ double entries[] = {
+ mean_i - mean_j,
+ std_err,
+ sig,
+ (mean_i - mean_j) - half_range,
+ (mean_i - mean_j) + half_range,
+ };
+ for (size_t k = 0; k < sizeof entries / sizeof *entries; k++)
+ pivot_table_put4 (table, k, j, i, test_idx,
+ pivot_value_new_number (entries[k]));
+ }
}
-
- if ( v > 0 )
- tab_hline (t, TAL_1, 0, n_cols - 1, (v * lines_per_variable) + 1);
}
- tab_submit (t);
+ pivot_table_submit (table);
}
-
-