-/* PSPP - computes sample statistics.
+/* PSPP - computes sample statistics. -*-c-*-
+
Copyright (C) 1997-9, 2000 Free Software Foundation, Inc.
- Written by Ben Pfaff <blp@gnu.org>.
+ Written by John Williams <johnr.williams@stonebow.otago.ac.nz>.
+ Almost completly re-written by John Darrington 2004
This program is free software; you can redistribute it and/or
modify it under the terms of the GNU General Public License as
02111-1307, USA. */
#include <config.h>
-#include <assert.h>
+#include <gsl/gsl_cdf.h>
+#include "error.h"
#include <stdio.h>
#include <stdlib.h>
#include <math.h>
#include "alloc.h"
#include "str.h"
-#include "dcdflib/cdflib.h"
#include "command.h"
#include "lexer.h"
#include "error.h"
#include "magic.h"
+#include "misc.h"
+#include "tab.h"
+#include "som.h"
+#include "value-labels.h"
#include "var.h"
#include "vfm.h"
+#include "hash.h"
+#include "t-test.h"
+#include "casefile.h"
+#include "levene.h"
/* (specification)
"T-TEST" (tts_):
- groups=custom;
+ +groups=custom;
+ +testval=double;
variables=varlist("PV_NO_SCRATCH | PV_NUMERIC");
- *+pairs=custom;
+ pairs=custom;
+missing=miss:!analysis/listwise,
incl:include/!exclude;
- +format=fmt:!labels/nolabels;
- +criteria=:ci(d:criteria,"%s > 0. && %s < 1.").
+ format=fmt:!labels/nolabels;
+ criteria=:cin(d:criteria,"%s > 0. && %s < 1.").
*/
/* (declarations) */
/* (functions) */
-#undef DEBUGGING
-#define DEBUGGING 1
-#include "debug-print.h"
-/* Command parsing information. */
-static struct cmd_t_test cmd;
+
+
+/* Function to use for testing for missing values */
+static is_missing_func value_is_missing;
/* Variable for the GROUPS subcommand, if given. */
-static struct variable *groups;
+static struct variable *indep_var;
/* GROUPS: Number of values specified by the user; the values
specified if any. */
-static int n_groups_values;
+
+static int n_group_values;
static union value groups_values[2];
+static enum comparison criteria[2];
-/* PAIRED: Number of pairs; each pair. */
-static int n_pairs;
-static struct variable *(*pairs)[2];
-
-/* Routines to scan data and perform t-tests */
-static void precalc (void);
-static void postcalc (void);
-static void g_postcalc (void);
-static void t_pairs (void);
-static void t_groups (void);
-static int groups_calc (struct ccase *);
-static int pairs_calc (struct ccase *);
-static int z_calc (struct ccase *);
-
-struct value_list
- {
- double sum;
- double ss;
- double n;
- struct value_list *next;
- };
-
-/* general workhorses - should move these to a separate library... */
-double variance (double n, double ss, double sum);
-
-double covariance (double x_sum, double x_n,
- double y_sum, double y_n, double ss);
-
-double pooled_variance (double n_1, double var_1,
- double n_2, double var_2);
-
-double oneway (double *f, double *p, struct value_list *list);
-
-double pearson_r (double c_xy, double c_xx, double c_yy);
-
-double f_sig (double f, double dfn, double dfd);
-double t_crt (double df, double q);
-double t_sig (double t, double df);
-
-/* massive function simply to remove any responsibility for output
- from the function which does the actual t-test calculations */
-void print_t_groups (struct variable * grps, union value * g1, union value * g2,
- double n1, double n2, double mean1, double mean2,
- double sd1, double sd2, double se1, double se2,
- double diff, double l_f, double l_p,
- double p_t, double p_sig, double p_df, double p_sed,
- double p_l, double p_h,
- double s_t, double s_sig, double s_df, double s_sed,
- double s_l, double s_h);
-
-/* Global variables to communicate between calc() and postcalc()
- should move to a structure in the p union of variable... */
-static double v1_n, v1_ss, v1_sum, v1_se, v1_var, v1_mean;
-static double v2_n, v2_ss, v2_sum, v2_se, v2_var, v2_mean;
-static double v1_z_sum, v1_z_ss;
-static double v2_z_sum, v2_z_ss;
-static double diff, se_diff, sp, xy_sum, xy_diff, xy_ss;
-static int cur_var;
-
-/* some defines for CDFlib */
-#define FIND_P 1
-#define FIND_CRITICAL_VALUE 2
-#define ERROR_SIG -1
-
-#ifdef DEBUGGING
-static void debug_print (void);
-#endif
-
-/* Parses and executes the T-TEST procedure. */
-int
-cmd_t_test (void)
+
+
+/* PAIRS: Number of pairs to be compared ; each pair. */
+static int n_pairs = 0 ;
+struct pair
{
- struct cmd_t_test cmd;
-
- if (!lex_force_match_id ("T"))
- return CMD_FAILURE;
- lex_match ('-');
- lex_match_id ("TEST");
+ /* The variables comprising the pair */
+ struct variable *v[2];
- if (!parse_t_test (&cmd))
- return CMD_FAILURE;
+ /* The number of valid variable pairs */
+ double n;
-#if DEBUGGING
- debug_print ();
-#endif
+ /* The sum of the members */
+ double sum[2];
- if (n_pairs > 0)
- procedure (precalc, pairs_calc, postcalc);
- else
- /* probably groups then... */
- {
- printf ("\n\n t-tests for independent samples of %s %s\n",
- groups->name, groups->label);
+ /* sum of squares of the members */
+ double ssq[2];
- for (cur_var = 0; cur_var < cmd.n_variables; cur_var++)
- {
- v1_n = v1_ss = v1_sum = v1_se = v1_var = v1_mean = 0.0;
- v2_n = v2_ss = v2_sum = v2_se = v2_var = v2_mean = 0.0;
- v1_z_sum = v1_z_ss = v2_z_sum = v2_z_ss = 0.0;
- diff = se_diff = sp = xy_diff = xy_ss = xy_sum = 0.0;
+ /* Std deviation of the members */
+ double std_dev[2];
- procedure (precalc, groups_calc, g_postcalc);
- procedure (precalc, z_calc, postcalc);
- }
- }
- return CMD_SUCCESS;
-}
+ /* Sample Std deviation of the members */
+ double s_std_dev[2];
-void
-precalc (void)
-{
- return; /* rilly void... */
-}
+ /* The means of the members */
+ double mean[2];
-int
-groups_calc (struct ccase * c)
-{
- int bad_weight;
- double group, w;
- struct variable *v = cmd.v_variables[cur_var];
- double X = c->data[v->fv].f;
-
- /* Get the weight for this case. */
- if (default_dict.weight_index == -1)
- w = 1.0;
- else
- {
- w = c->data[default_dict.weight_index].f;
- if (w <= 0.0 || w == SYSMIS)
- {
- w = 0.0;
- bad_weight = 1;
- printf ("Bad weight\n");
- }
- }
+ /* The correlation coefficient between the variables */
+ double correlation;
- if (X == SYSMIS || X == 0.0) /* FIXME: should be USER_MISSING? */
- {
- /* printf("Missing value\n"); */
- return 1;
- }
- else
- {
- X = X * w;
- group = c->data[groups->fv].f;
+ /* The sum of the differences */
+ double sum_of_diffs;
- if (group == groups_values[0].f)
- {
- v1_sum += X;
- v1_ss += X * X;
- v1_n += w;
- }
- else if (group == groups_values[1].f)
- {
- v2_sum += X;
- v2_ss += X * X;
- v2_n += w;
- }
- }
+ /* The sum of the products */
+ double sum_of_prod;
- return 1;
-}
+ /* The mean of the differences */
+ double mean_diff;
-void
-g_postcalc (void)
-{
- v1_mean = v1_sum / v1_n;
- v2_mean = v2_sum / v2_n;
- return;
-}
+ /* The sum of the squares of the differences */
+ double ssq_diffs;
+
+ /* The std deviation of the differences */
+ double std_dev_diff;
+};
+
+static struct pair *pairs=0;
+
+static int parse_value (union value * v, int type) ;
-int /* this pass generates the z-zcores */
-z_calc (struct ccase * c)
+/* Structures and Functions for the Statistics Summary Box */
+struct ssbox;
+typedef void populate_ssbox_func(struct ssbox *ssb,
+ struct cmd_t_test *cmd);
+typedef void finalize_ssbox_func(struct ssbox *ssb);
+
+struct ssbox
{
- int bad_weight;
- double group, z, w;
- struct variable *v = cmd.v_variables[cur_var];
- double X = c->data[v->fv].f;
+ struct tab_table *t;
- z = 0.0;
+ populate_ssbox_func *populate;
+ finalize_ssbox_func *finalize;
- /* Get the weight for this case. */
- if (default_dict.weight_index == -1)
- w = 1.0;
- else
- {
- w = c->data[default_dict.weight_index].f;
- if (w <= 0.0 || w == SYSMIS)
- {
- w = 0.0;
- bad_weight = 1;
- }
- }
+};
- if (X == SYSMIS || X == 0.0) /* FIXME: how to specify user missing? */
- {
- return 1;
- }
- else
- {
- group = c->data[groups->fv].f;
- X = w * X;
+/* Create a ssbox */
+void ssbox_create(struct ssbox *ssb, struct cmd_t_test *cmd, int mode);
- if (group == groups_values[0].f)
- {
- z = fabs (X - v1_mean);
- v1_z_sum += z;
- v1_z_ss += pow (z, 2);
- }
- else if (group == groups_values[1].f)
- {
- z = fabs (X - v2_mean);
- v2_z_ss += pow (z, 2);
- v2_z_sum += z;
- }
- }
+/* Populate a ssbox according to cmd */
+void ssbox_populate(struct ssbox *ssb, struct cmd_t_test *cmd);
- return 1;
-}
+/* Submit and destroy a ssbox */
+void ssbox_finalize(struct ssbox *ssb);
+
+/* A function to create, populate and submit the Paired Samples Correlation
+ box */
+void pscbox(void);
+
+
+/* Structures and Functions for the Test Results Box */
+struct trbox;
+
+typedef void populate_trbox_func(struct trbox *trb,
+ struct cmd_t_test *cmd);
+typedef void finalize_trbox_func(struct trbox *trb);
+
+struct trbox {
+ struct tab_table *t;
+ populate_trbox_func *populate;
+ finalize_trbox_func *finalize;
+};
+
+/* Create a trbox */
+void trbox_create(struct trbox *trb, struct cmd_t_test *cmd, int mode);
+
+/* Populate a ssbox according to cmd */
+void trbox_populate(struct trbox *trb, struct cmd_t_test *cmd);
+
+/* Submit and destroy a ssbox */
+void trbox_finalize(struct trbox *trb);
+
+/* Which mode was T-TEST invoked */
+enum {
+ T_1_SAMPLE = 0 ,
+ T_IND_SAMPLES,
+ T_PAIRED
+};
+
+
+static int common_calc (const struct ccase *, void *);
+static void common_precalc (struct cmd_t_test *);
+static void common_postcalc (struct cmd_t_test *);
+
+static int one_sample_calc (const struct ccase *, void *);
+static void one_sample_precalc (struct cmd_t_test *);
+static void one_sample_postcalc (struct cmd_t_test *);
+
+static int paired_calc (const struct ccase *, void *);
+static void paired_precalc (struct cmd_t_test *);
+static void paired_postcalc (struct cmd_t_test *);
+
+static void group_precalc (struct cmd_t_test *);
+static int group_calc (const struct ccase *, struct cmd_t_test *);
+static void group_postcalc (struct cmd_t_test *);
+static int compare_var_name (const void *a_, const void *b_, void *v_ UNUSED);
+static unsigned hash_var_name (const void *a_, void *v_ UNUSED);
+
+static void calculate(const struct casefile *cf, void *_mode);
+
+static int mode;
+
+static struct cmd_t_test cmd;
+
+static int bad_weight_warn;
+
int
-pairs_calc (struct ccase * c)
+cmd_t_test(void)
{
- int i;
- struct variable *v1, *v2;
- double X, Y;
- for (i = 0; i < n_pairs; i++)
- {
- v1 = pairs[i][0];
- v2 = pairs[i][1];
- X = c->data[v1->fv].f;
- Y = c->data[v2->fv].f;
- if (X == SYSMIS || Y == SYSMIS)
+ if ( !parse_t_test(&cmd) )
+ return CMD_FAILURE;
+
+ if (! cmd.sbc_criteria)
+ cmd.criteria=0.95;
+
+ {
+ int m=0;
+ if (cmd.sbc_testval) ++m;
+ if (cmd.sbc_groups) ++m;
+ if (cmd.sbc_pairs) ++m;
+
+ if ( m != 1)
+ {
+ msg(SE,
+ _("TESTVAL, GROUPS and PAIRS subcommands are mutually exclusive.")
+ );
+ return CMD_FAILURE;
+ }
+ }
+
+ if (cmd.sbc_testval)
+ mode=T_1_SAMPLE;
+ else if (cmd.sbc_groups)
+ mode=T_IND_SAMPLES;
+ else
+ mode=T_PAIRED;
+
+ if ( mode == T_PAIRED)
+ {
+ if (cmd.sbc_variables)
{
- printf ("Missing value\n");
+ msg(SE, _("VARIABLES subcommand is not appropriate with PAIRS"));
+ return CMD_FAILURE;
}
else
{
- xy_sum += X * Y;
- xy_diff += (X - Y);
- xy_ss += pow ((X - Y), 2);
- v1_sum += X;
- v2_sum += Y;
- v1_n++;
- v2_n++;
- v1_ss += (X * X);
- v2_ss += (Y * Y);
- }
- }
+ /* Iterate through the pairs and put each variable that is a
+ member of a pair into cmd.v_variables */
- return 1;
-}
+ int i;
+ struct hsh_iterator hi;
+ struct hsh_table *hash;
+ struct variable *v;
-void
-postcalc (void)
-{
- /* Calculate basic statistics */
- v1_var = variance (v1_n, v1_ss, v1_sum); /* variances */
- v2_var = variance (v2_n, v2_ss, v2_sum);
- v1_se = sqrt (v1_var / v1_n); /* standard errors */
- v2_se = sqrt (v2_var / v2_n);
- diff = v1_mean - v2_mean;
-
- if (n_pairs > 0)
+ hash=hsh_create(n_pairs,compare_var_name,hash_var_name,0,0);
+
+ for (i=0; i < n_pairs; ++i)
+ {
+ hsh_insert(hash,pairs[i].v[0]);
+ hsh_insert(hash,pairs[i].v[1]);
+ }
+
+ assert(cmd.n_variables == 0);
+ cmd.n_variables = hsh_count(hash);
+
+ cmd.v_variables = xrealloc(cmd.v_variables,
+ sizeof(struct variable) * cmd.n_variables);
+ /* Iterate through the hash */
+ for (i=0,v = (struct variable *) hsh_first(hash,&hi);
+ v != 0;
+ v=hsh_next(hash,&hi) )
+ cmd.v_variables[i++]=v;
+
+ hsh_destroy(hash);
+ }
+ }
+ else if ( !cmd.sbc_variables)
{
- t_pairs ();
+ msg(SE, _("One or more VARIABLES must be specified."));
+ return CMD_FAILURE;
}
+
+
+ /* If /MISSING=INCLUDE is set, then user missing values are ignored */
+ if (cmd.incl == TTS_INCLUDE )
+ value_is_missing = is_system_missing;
else
+ value_is_missing = is_missing;
+
+ bad_weight_warn = 1;
+
+ multipass_procedure_with_splits (calculate, &cmd);
+
+ n_pairs=0;
+ free(pairs);
+ pairs=0;
+
+ if ( mode == T_IND_SAMPLES)
{
- t_groups ();
+ int i;
+ /* Destroy any group statistics we created */
+ for (i= 0 ; i < cmd.n_variables ; ++i )
+ {
+ free(cmd.v_variables[i]->p.t_t.gs);
+ }
}
-
- return;
+
+ return CMD_SUCCESS;
}
-void
-t_groups (void)
+static int
+tts_custom_groups (struct cmd_t_test *cmd UNUSED)
{
- double df_pooled, t_pooled, t_sep, p_pooled, p_sep;
- double crt_t_p, crt_t_s, tmp, v1_z, v2_z, f_levene, p_levene;
- double df_sep, se_diff_s, se_diff_p;
- struct value_list *val_1, *val_2;
-
- /* Levene's test */
- val_1 = malloc (sizeof (struct value_list));
- val_1->sum = v1_z_sum;
- val_1->ss = v1_z_ss;
- val_1->n = v1_n;
- val_2 = malloc (sizeof (struct value_list));
- val_2->sum = v2_z_sum;
- val_2->ss = v2_z_ss;
- val_2->n = v2_n;
-
- val_1->next = val_2;
- val_2->next = NULL;
-
- f_levene = oneway (&f_levene, &p_levene, val_1);
-
- /* T test results for pooled variances */
- se_diff_p = sqrt (pooled_variance (v1_n, v1_var, v2_n, v2_var));
- df_pooled = v1_n + v2_n - 2.0;
- t_pooled = diff / se_diff_p;
- p_pooled = t_sig (t_pooled, df_pooled);
- crt_t_p = t_crt (df_pooled, 0.025);
-
- if ((2.0 * p_pooled) >= 1.0)
- p_pooled = 1.0 - p_pooled;
-
- /* oh god, the separate variance calculations... */
- t_sep = diff / sqrt ((v1_var / v1_n) + (v2_var / v2_n));
-
- tmp = (v1_var / v1_n) + (v2_var / v2_n);
- tmp = (v1_var / v1_n) / tmp;
- tmp = pow (tmp, 2);
- tmp = tmp / (v1_n - 1.0);
- v1_z = tmp;
-
- tmp = (v1_var / v1_n) + (v2_var / v2_n);
- tmp = (v2_var / v2_n) / tmp;
- tmp = pow (tmp, 2);
- tmp = tmp / (v2_n - 1.0);
- v2_z = tmp;
-
- tmp = 1.0 / (v1_z + v2_z);
-
- df_sep = tmp;
- p_sep = t_sig (t_sep, df_sep);
- if ((2.0 * p_sep) >= 1.0)
- p_sep = 1.0 - p_sep;
- crt_t_s = t_crt (df_sep, 0.025);
- se_diff_s = sqrt ((v1_var / v1_n) + (v2_var / v2_n));
-
- /* FIXME: convert to a proper PSPP output call */
- print_t_groups (groups, &groups_values[0], &groups_values[1],
- v1_n, v2_n, v1_mean, v2_mean,
- sqrt (v1_var), sqrt (v2_var), v1_se, v2_se,
- diff, f_levene, p_levene,
- t_pooled, 2.0 * p_pooled, df_pooled, se_diff_p,
- diff - (crt_t_p * se_diff_p), diff + (crt_t_p * se_diff_p),
- t_sep, 2.0 * p_sep, df_sep, se_diff_s,
- diff - (crt_t_s * se_diff_s), diff + (crt_t_s * se_diff_s));
- return;
-}
-void
-t_pairs (void)
-{
- double cov12, cov11, cov22, r, t, p, crt_t, sp, r_t, r_p;
- struct variable *v1, *v2;
-
- v1 = pairs[0][0];
- v2 = pairs[0][1];
- cov12 = covariance (v1_sum, v1_n, v2_sum, v2_n, xy_sum);
- cov11 = covariance (v1_sum, v1_n, v1_sum, v1_n, v1_ss);
- cov22 = covariance (v2_sum, v2_n, v2_sum, v2_n, v2_ss);
- r = pearson_r (cov12, cov11, cov22);
- /* this t and it's associated p is a significance test for the pearson's r */
- r_t = r * sqrt ((v1_n - 2.0) / (1.0 - (r * r)));
- r_p = t_sig (r_t, v1_n - 2.0);
-
- /* now we move to the t test for the difference in means */
- diff = xy_diff / v1_n;
- sp = sqrt (variance (v1_n, xy_ss, xy_diff));
- se_diff = sp / sqrt (v1_n);
- t = diff / se_diff;
- crt_t = t_crt (v1_n - 1.0, 0.025);
- p = t_sig (t, v1_n - 1.0);
-
-
- printf (" Number of 2-tail\n");
- printf (" Variable pairs Corr Sig Mean SD SE of Mean\n");
- printf ("---------------------------------------------------------------\n");
- printf ("%s %8.4f %8.4f %8.4f\n",
- v1->name, v1_mean, sqrt (v1_var), v1_se);
- printf (" %8.4f %0.4f %0.4f\n", v1_n, r, r_p);
- printf ("%s %8.4f %8.4f %8.4f\n",
- v2->name, v2_mean, sqrt (v2_var), v2_se);
- printf ("---------------------------------------------------------------\n");
-
- printf ("\n\n\n");
- printf (" Paired Differences |\n");
- printf (" Mean SD SE of Mean | t-value df 2-tail Sig\n");
- printf ("--------------------------------------|---------------------------\n");
-
- printf ("%8.4f %8.4f %8.4f | %8.4f %8.4f %8.4f\n",
- diff, sp, se_diff, t, v1_n - 1.0, 2.0 * (1.0 - p));
-
- printf ("95pc CI (%8.4f, %8.4f) |\n\n",
- diff - (se_diff * crt_t), diff + (se_diff * crt_t));
-
- return;
-}
+ lex_match('=');
-static int parse_value (union value *);
+ if (token != T_ALL &&
+ (token != T_ID || dict_lookup_var (default_dict, tokid) == NULL)
+ )
+ {
+ msg(SE,_("`%s' is not a variable name"),tokid);
+ return 0;
+ }
-/* Parses the GROUPS subcommand. */
-int
-tts_custom_groups (struct cmd_t_test *cmd unused)
-{
- groups = parse_variable ();
- if (!groups)
+ indep_var = parse_variable ();
+ if (!indep_var)
{
- lex_error (_("expecting variable name in GROUPS subcommand"));
+ lex_error ("expecting variable name in GROUPS subcommand");
return 0;
}
- if (groups->type == T_STRING && groups->width > MAX_SHORT_STRING)
+
+ if (indep_var->type == T_STRING && indep_var->width > MAX_SHORT_STRING)
{
msg (SE, _("Long string variable %s is not valid here."),
- groups->name);
+ indep_var->name);
return 0;
}
if (!lex_match ('('))
{
- if (groups->type == NUMERIC)
+ if (indep_var->type == NUMERIC)
{
- n_groups_values = 2;
groups_values[0].f = 1;
groups_values[1].f = 2;
+ criteria[0] = criteria[1] = CMP_EQ;
+ n_group_values = 2;
return 1;
}
else
return 0;
}
}
-
- if (!parse_value (&groups_values[0]))
- return 0;
- n_groups_values = 1;
- lex_match (',');
+ if (!parse_value (&groups_values[0],indep_var->type))
+ return 0;
+ lex_match (',');
if (lex_match (')'))
- return 1;
+ {
+ criteria[0] = CMP_LE;
+ criteria[1] = CMP_GT;
+ groups_values[1] = groups_values[0];
+ n_group_values = 1;
+ return 1;
+ }
- if (!parse_value (&groups_values[1]))
+ if (!parse_value (&groups_values[1],indep_var->type))
return 0;
- n_groups_values = 2;
-
+
+ n_group_values = 2;
if (!lex_force_match (')'))
return 0;
+ criteria[0] = criteria[1] = CMP_EQ;
return 1;
}
-/* Parses the current token (numeric or string, depending on the
- variable in `groups') into value V and returns success. */
-static int
-parse_value (union value * v)
-{
- if (groups->type == NUMERIC)
- {
- if (!lex_force_num ())
- return 0;
- v->f = tokval;
- }
- else
- {
- if (!lex_force_string ())
- return 0;
- strncpy (v->s, ds_value (&tokstr), ds_length (&tokstr));
- }
-
- lex_get ();
-
- return 1;
-}
-/* Parses the PAIRS subcommand. */
static int
-tts_custom_pairs (struct cmd_t_test *cmd unused)
+tts_custom_pairs (struct cmd_t_test *cmd UNUSED)
{
struct variable **vars;
- int n_before_WITH;
int n_vars;
- int paired;
- int extra;
-#if DEBUGGING
- int n_predicted;
-#endif
-
- if ((token != T_ID || !is_varname (tokid)) && token != T_ALL)
- return 2;
- if (!parse_variables (&default_dict, &vars, &n_vars,
- PV_DUPLICATE | PV_NUMERIC | PV_NO_SCRATCH))
- return 0;
+ int n_pairs_local;
+
+ int n_before_WITH ;
+ int n_after_WITH = -1;
+ int paired ; /* Was the PAIRED keyword given ? */
+
+ lex_match('=');
+
+ if ((token != T_ID || dict_lookup_var (default_dict, tokid) == NULL)
+ && token != T_ALL)
+ {
+ msg(SE,_("`%s' is not a variable name"),tokid);
+ return 0;
+ }
+ n_vars=0;
+ if (!parse_variables (default_dict, &vars, &n_vars,
+ PV_DUPLICATE | PV_NUMERIC | PV_NO_SCRATCH))
+ {
+ free (vars);
+ return 0;
+ }
assert (n_vars);
+
+ n_before_WITH=0;
if (lex_match (T_WITH))
{
n_before_WITH = n_vars;
-
- if (!parse_variables (&default_dict, &vars, &n_vars,
+ if (!parse_variables (default_dict, &vars, &n_vars,
PV_DUPLICATE | PV_APPEND
| PV_NUMERIC | PV_NO_SCRATCH))
{
free (vars);
return 0;
}
+ n_after_WITH = n_vars - n_before_WITH;
}
- else
- n_before_WITH = 0;
paired = (lex_match ('(') && lex_match_id ("PAIRED") && lex_match (')'));
+ /* Determine the number of pairs needed */
if (paired)
{
- if (n_before_WITH * 2 != n_vars)
+ if (n_before_WITH != n_after_WITH)
{
free (vars);
msg (SE, _("PAIRED was specified but the number of variables "
"preceding WITH (%d) did not match the number "
"following (%d)."),
- n_before_WITH, n_vars - n_before_WITH);
+ n_before_WITH, n_after_WITH );
return 0;
}
-
- extra = n_before_WITH;
+ n_pairs_local=n_before_WITH;
}
- else if (n_before_WITH)
- extra = n_before_WITH * (n_vars - n_before_WITH);
- else
+ else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
+ {
+ n_pairs_local=n_before_WITH * n_after_WITH ;
+ }
+ else /* Neither WITH nor PAIRED keyword given */
{
if (n_vars < 2)
{
return 0;
}
- extra = n_vars * (n_vars - 1) / 2;
+ /* how many ways can you pick 2 from n_vars ? */
+ n_pairs_local = n_vars * (n_vars -1 ) /2 ;
}
-#if DEBUGGING
- n_predicted = n_pairs + extra;
-#endif
- pairs = xrealloc (pairs, sizeof (struct variable *[2]) * (n_pairs + extra));
+ /* Allocate storage for the pairs */
+ pairs = xrealloc(pairs, sizeof(struct pair) * (n_pairs + n_pairs_local) );
- if (paired)
+ /* Populate the pairs with the appropriate variables */
+ if ( paired )
{
int i;
- for (i = 0; i < extra; i++)
+ assert(n_pairs_local == n_vars/2);
+ for (i = 0; i < n_pairs_local ; ++i)
{
- pairs[n_pairs][0] = vars[i];
- pairs[n_pairs++][1] = vars[i + extra];
+ pairs[i].v[n_pairs+0] = vars[i];
+ pairs[i].v[n_pairs+1] = vars[i+n_pairs_local];
}
}
- else if (n_before_WITH)
+ else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
{
- int i;
+ int i,j;
+ int p=n_pairs;
- for (i = 0; i < n_before_WITH; i++)
+ for(i=0 ; i < n_before_WITH ; ++i )
{
- int j;
-
- for (j = n_before_WITH; j < n_vars; j++)
+ for(j=0 ; j < n_after_WITH ; ++j)
{
- pairs[n_pairs][0] = vars[i];
- pairs[n_pairs++][1] = vars[j];
+ pairs[p].v[0] = vars[i];
+ pairs[p].v[1] = vars[j+n_before_WITH];
+ ++p;
}
}
}
- else
+ else /* Neither WITH nor PAIRED given */
{
- int i;
-
- for (i = 0; i < n_vars; i++)
+ int i,j;
+ int p=n_pairs;
+
+ for(i=0 ; i < n_vars ; ++i )
{
- int j;
-
- for (j = i + 1; j < n_vars; j++)
+ for(j=i+1 ; j < n_vars ; ++j)
{
- pairs[n_pairs][0] = vars[i];
- pairs[n_pairs++][1] = vars[j];
+ pairs[p].v[0] = vars[i];
+ pairs[p].v[1] = vars[j];
+ ++p;
}
}
}
-#if DEBUGGING
- assert (n_pairs == n_predicted);
-#endif
+ n_pairs+=n_pairs_local;
- free (vars);
return 1;
}
-#if DEBUGGING
-static void
-debug_print (void)
+/* Parses the current token (numeric or string, depending on type)
+ value v and returns success. */
+static int
+parse_value (union value * v, int type )
{
- printf ("T-TEST\n");
- if (groups)
+ if (type == NUMERIC)
{
- printf (" GROUPS=%s", groups->name);
- if (n_groups_values)
- {
- int i;
-
- printf (" (");
- for (i = 0; i < n_groups_values; i++)
- if (groups->type == NUMERIC)
- printf ("%g%s", groups_values[i].f, i ? " " : "");
- else
- printf ("%.*s%s", groups->width, groups_values[i].s,
- i ? " " : "");
- printf (")");
- }
- printf ("\n");
+ if (!lex_force_num ())
+ return 0;
+ v->f = tokval;
}
- if (cmd.n_variables)
+ else
{
- int i;
-
- printf (" VARIABLES=");
- for (i = 0; i < cmd.n_variables; i++)
- printf ("%s ", cmd.v_variables[i]->name);
- printf ("\n");
+ if (!lex_force_string ())
+ return 0;
+ strncpy (v->s, ds_c_str (&tokstr), ds_length (&tokstr));
}
- if (cmd.sbc_pairs)
- {
- int i;
- printf (" PAIRS=");
- for (i = 0; i < n_pairs; i++)
- printf ("%s ", pairs[i][0]->name);
- printf ("WITH");
- for (i = 0; i < n_pairs; i++)
- printf (" %s", pairs[i][1]->name);
- printf (" (PAIRED)\n");
- }
- printf (" MISSING=%s %s\n",
- cmd.miss == TTS_ANALYSIS ? "ANALYSIS" : "LISTWISE",
- cmd.miss == TTS_INCLUDE ? "INCLUDE" : "EXCLUDE");
- printf (" FORMAT=%s\n",
- cmd.fmt == TTS_LABELS ? "LABELS" : "NOLABELS");
- if (cmd.criteria != NOT_LONG)
- printf (" CRITERIA=%f\n", cmd.criteria);
+ lex_get ();
+
+ return 1;
+}
+
+
+/* Implementation of the SSBOX object */
+
+void ssbox_base_init(struct ssbox *this, int cols,int rows);
+
+void ssbox_base_finalize(struct ssbox *ssb);
+
+void ssbox_one_sample_init(struct ssbox *this,
+ struct cmd_t_test *cmd );
+
+void ssbox_independent_samples_init(struct ssbox *this,
+ struct cmd_t_test *cmd);
+
+void ssbox_paired_init(struct ssbox *this,
+ struct cmd_t_test *cmd);
+
+/* Factory to create an ssbox */
+void
+ssbox_create(struct ssbox *ssb, struct cmd_t_test *cmd, int mode)
+{
+ switch (mode)
+ {
+ case T_1_SAMPLE:
+ ssbox_one_sample_init(ssb,cmd);
+ break;
+ case T_IND_SAMPLES:
+ ssbox_independent_samples_init(ssb,cmd);
+ break;
+ case T_PAIRED:
+ ssbox_paired_init(ssb,cmd);
+ break;
+ default:
+ assert(0);
+ }
}
-#endif /* DEBUGGING */
-/* Here are some general routines tha should probably be moved into
- a separate library and documented as part of the PSPP "API" */
-double
-variance (double n, double ss, double sum)
+/* Despatcher for the populate method */
+void
+ssbox_populate(struct ssbox *ssb,struct cmd_t_test *cmd)
+{
+ ssb->populate(ssb,cmd);
+}
+
+
+/* Despatcher for finalize */
+void
+ssbox_finalize(struct ssbox *ssb)
{
- return ((ss - ((sum * sum) / n)) / (n - 1.0));
+ ssb->finalize(ssb);
}
-double
-pooled_variance (double n_1, double var_1, double n_2, double var_2)
+
+/* Submit the box and clear up */
+void
+ssbox_base_finalize(struct ssbox *ssb)
{
- double tmp;
+ tab_submit(ssb->t);
+}
- tmp = n_1 + n_2 - 2.0;
- tmp = (((n_1 - 1.0) * var_1) + ((n_2 - 1.0) * var_2)) / tmp;
- tmp = tmp * ((n_1 + n_2) / (n_1 * n_2));
- return tmp;
+/* Initialize a ssbox struct */
+void
+ssbox_base_init(struct ssbox *this, int cols,int rows)
+{
+ this->finalize = ssbox_base_finalize;
+ this->t = tab_create (cols, rows, 0);
+
+ tab_columns (this->t, SOM_COL_DOWN, 1);
+ tab_headers (this->t,0,0,1,0);
+ tab_box (this->t, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols -1, rows -1 );
+ tab_hline(this->t, TAL_2,0,cols-1,1);
+ tab_dim (this->t, tab_natural_dimensions);
}
-double
-oneway (double *f, double *p, struct value_list *levels)
+void ssbox_one_sample_populate(struct ssbox *ssb,
+ struct cmd_t_test *cmd);
+
+/* Initialize the one_sample ssbox */
+void
+ssbox_one_sample_init(struct ssbox *this,
+ struct cmd_t_test *cmd )
{
- double k, SSTR, SSE, SSTO, N, MSTR, MSE, sum, dftr, dfe, print;
- struct value_list *g;
+ const int hsize=5;
+ const int vsize=cmd->n_variables+1;
+
+ this->populate = ssbox_one_sample_populate;
+
+ ssbox_base_init(this, hsize,vsize);
+ tab_title (this->t, 0, _("One-Sample Statistics"));
+ tab_vline(this->t, TAL_2, 1,0,vsize);
+ tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, _("N"));
+ tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
+ tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
+ tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
+}
- k = 0.0;
+void ssbox_independent_samples_populate(struct ssbox *ssb,
+ struct cmd_t_test *cmd);
+
+/* Initialize the independent samples ssbox */
+void
+ssbox_independent_samples_init(struct ssbox *this,
+ struct cmd_t_test *cmd)
+{
+ int hsize=6;
+ int vsize = cmd->n_variables*2 +1;
+
+ this->populate = ssbox_independent_samples_populate;
+
+ ssbox_base_init(this, hsize,vsize);
+ tab_title (this->t, 0, _("Group Statistics"));
+ tab_vline(this->t,0,1,0,vsize);
+ tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, indep_var->name);
+ tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("N"));
+ tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
+ tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
+ tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
+}
- for (g = levels; g != NULL; g = g->next)
- {
- k++;
- sum += g->sum;
- N += g->n;
- SSTR += g->ss - (pow (g->sum, 2) / g->n);
- SSTO += g->ss;
- }
- SSTO = SSTO - (pow (sum, 2) / N);
- SSE = SSTO - SSTR;
+/* Populate the ssbox for independent samples */
+void
+ssbox_independent_samples_populate(struct ssbox *ssb,
+ struct cmd_t_test *cmd)
+{
+ int i;
- dftr = N - k;
- dfe = k - 1.0;
- MSTR = SSTR / dftr;
- MSE = SSE / dfe;
+ char *val_lab0=0;
+ char *val_lab1=0;
- *f = (MSE / MSTR);
- *p = f_sig (*f, dfe, dftr);
+ char prefix[2][3]={"",""};
- print = 1.0;
- if (print == 1.0)
+ if ( indep_var->type == NUMERIC )
{
- printf ("sum1 %f, sum2 %f, ss1 %f, ss2 %f\n",
- levels->sum, levels->next->sum, levels->ss, levels->next->ss);
- printf (" - - - - - - O N E W A Y - - - - - -\n\n");
- printf (" Variable %s %s\n",
- cmd.v_variables[0]->name, cmd.v_variables[0]->label);
- printf ("By Variable %s %s\n", groups->name, groups->label);
- printf ("\n Analysis of Variance\n\n");
- printf (" Sum of Mean F F\n");
- printf ("Source D.F. Squares Squares Ratio Prob\n\n");
- printf ("Between %8.0f %8.4f %8.4f %8.4f %8.4f\n",
- dfe, SSE, MSE, *f, *p);
- printf ("Within %8.0f %8.4f %8.4f\n", dftr, SSTR, MSTR);
- printf ("Total %8.0f %8.4f\n\n\n", N - 1.0, SSTO);
+ val_lab0 = val_labs_find( indep_var->val_labs,groups_values[0]);
+ val_lab1 = val_labs_find( indep_var->val_labs,groups_values[1]);
+ }
+ else
+ {
+ val_lab0 = groups_values[0].s;
+ val_lab1 = groups_values[1].s;
}
- return (*f);
-}
-double
-f_sig (double f, double dfn, double dfd)
-{
- int which, status;
- double p, q, bound;
+ if (n_group_values == 1)
+ {
+ strcpy(prefix[0],"< ");
+ strcpy(prefix[1],">=");
+ }
- which = FIND_P;
- status = 1;
- p = q = bound = 0.0;
- cdff (&which, &p, &q, &f, &dfn, &dfd, &status, &bound);
+ assert(ssb->t);
- switch (status)
+ for (i=0; i < cmd->n_variables; ++i)
{
- case -1:
- {
- printf ("Parameter 1 is out of range\n");
- break;
- }
- case -2:
- {
- printf ("Parameter 2 is out of range\n");
- break;
- }
- case -3:
- {
- printf ("Parameter 3 is out of range\n");
- break;
- }
- case -4:
- {
- printf ("Parameter 4 is out of range\n");
- break;
- }
- case -5:
- {
- printf ("Parameter 5 is out of range\n");
- break;
- }
- case -6:
- {
- printf ("Parameter 6 is out of range\n");
- break;
- }
- case -7:
- {
- printf ("Parameter 7 is out of range\n");
- break;
- }
- case -8:
- {
- printf ("Parameter 8 is out of range\n");
- break;
- }
- case 0:
+ int g;
+
+ tab_text (ssb->t, 0, i*2+1, TAB_LEFT, cmd->v_variables[i]->name);
+
+ if (val_lab0)
+ tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF,
+ "%s%s", prefix[0], val_lab0);
+ else
+ tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF,
+ "%s%g", prefix[0], groups_values[0].f);
+
+
+ if (val_lab1)
+ tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF,
+ "%s%s", prefix[1], val_lab1);
+ else
+ tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF,
+ "%s%g", prefix[1], groups_values[1].f);
+
+ /* Fill in the group statistics */
+ for ( g=0; g < 2 ; ++g )
+ {
+ struct group_statistics *gs = &cmd->v_variables[i]->p.t_t.gs[g];
+
+ tab_float(ssb->t, 2 ,i*2+g+1, TAB_RIGHT, gs->n, 2, 0);
+ tab_float(ssb->t, 3 ,i*2+g+1, TAB_RIGHT, gs->mean, 8, 2);
+ tab_float(ssb->t, 4 ,i*2+g+1, TAB_RIGHT, gs->std_dev, 8, 3);
+ tab_float(ssb->t, 5 ,i*2+g+1, TAB_RIGHT, gs->se_mean, 8, 3);
+ }
+ }
+}
+
+
+void ssbox_paired_populate(struct ssbox *ssb,
+ struct cmd_t_test *cmd);
+
+/* Initialize the paired values ssbox */
+void
+ssbox_paired_init(struct ssbox *this, struct cmd_t_test *cmd UNUSED)
+{
+ int hsize=6;
+
+ int vsize = n_pairs*2+1;
+
+ this->populate = ssbox_paired_populate;
+
+ ssbox_base_init(this, hsize,vsize);
+ tab_title (this->t, 0, _("Paired Sample Statistics"));
+ tab_vline(this->t,TAL_0,1,0,vsize-1);
+ tab_vline(this->t,TAL_2,2,0,vsize-1);
+ tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
+ tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("N"));
+ tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
+ tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
+}
+
+
+/* Populate the ssbox for paired values */
+void
+ssbox_paired_populate(struct ssbox *ssb,struct cmd_t_test *cmd UNUSED)
+{
+ int i;
+
+ assert(ssb->t);
+
+ for (i=0; i < n_pairs; ++i)
+ {
+ int j;
+
+ tab_text (ssb->t, 0, i*2+1, TAB_LEFT | TAT_PRINTF , _("Pair %d"),i);
+
+ for (j=0 ; j < 2 ; ++j)
+ {
+ struct group_statistics *gs;
+
+ gs=&pairs[i].v[j]->p.t_t.ugs;
+
+ /* Titles */
+
+ tab_text (ssb->t, 1, i*2+j+1, TAB_LEFT, pairs[i].v[j]->name);
+
+ /* Values */
+ tab_float (ssb->t,2, i*2+j+1, TAB_RIGHT, pairs[i].mean[j], 8, 2);
+ tab_float (ssb->t,3, i*2+j+1, TAB_RIGHT, pairs[i].n, 2, 0);
+ tab_float (ssb->t,4, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j], 8, 3);
+ tab_float (ssb->t,5, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j]/sqrt(pairs[i].n), 8, 3);
+
+ }
+ }
+}
+
+/* Populate the one sample ssbox */
+void
+ssbox_one_sample_populate(struct ssbox *ssb, struct cmd_t_test *cmd)
+{
+ int i;
+
+ assert(ssb->t);
+
+ for (i=0; i < cmd->n_variables; ++i)
+ {
+ struct group_statistics *gs;
+ gs= &cmd->v_variables[i]->p.t_t.ugs;
+
+ tab_text (ssb->t, 0, i+1, TAB_LEFT, cmd->v_variables[i]->name);
+ tab_float (ssb->t,1, i+1, TAB_RIGHT, gs->n, 2, 0);
+ tab_float (ssb->t,2, i+1, TAB_RIGHT, gs->mean, 8, 2);
+ tab_float (ssb->t,3, i+1, TAB_RIGHT, gs->std_dev, 8, 2);
+ tab_float (ssb->t,4, i+1, TAB_RIGHT, gs->se_mean, 8, 3);
+ }
+
+}
+
+
+
+/* Implementation of the Test Results box struct */
+
+void trbox_base_init(struct trbox *self,int n_vars, int cols);
+void trbox_base_finalize(struct trbox *trb);
+
+void trbox_independent_samples_init(struct trbox *trb,
+ struct cmd_t_test *cmd );
+
+void trbox_independent_samples_populate(struct trbox *trb,
+ struct cmd_t_test *cmd);
+
+void trbox_one_sample_init(struct trbox *self,
+ struct cmd_t_test *cmd );
+
+void trbox_one_sample_populate(struct trbox *trb,
+ struct cmd_t_test *cmd);
+
+void trbox_paired_init(struct trbox *self,
+ struct cmd_t_test *cmd );
+
+void trbox_paired_populate(struct trbox *trb,
+ struct cmd_t_test *cmd);
+
+
+
+/* Create a trbox according to mode*/
+void
+trbox_create(struct trbox *trb,
+ struct cmd_t_test *cmd, int mode)
+{
+ switch (mode)
{
- /* printf( "Command completed successfully\n" ); */
+ case T_1_SAMPLE:
+ trbox_one_sample_init(trb,cmd);
break;
- }
- case 1:
- {
- printf ("Answer appears to be lower than the lowest search bound\n");
+ case T_IND_SAMPLES:
+ trbox_independent_samples_init(trb,cmd);
break;
- }
- case 2:
- {
- printf ("Answer appears to be higher than the greatest search bound\n");
+ case T_PAIRED:
+ trbox_paired_init(trb,cmd);
break;
+ default:
+ assert(0);
}
- case 3:
+}
+
+/* Populate a trbox according to cmd */
+void
+trbox_populate(struct trbox *trb, struct cmd_t_test *cmd)
+{
+ trb->populate(trb,cmd);
+}
+
+/* Submit and destroy a trbox */
+void
+trbox_finalize(struct trbox *trb)
+{
+ trb->finalize(trb);
+}
+
+/* Initialize the independent samples trbox */
+void
+trbox_independent_samples_init(struct trbox *self,
+ struct cmd_t_test *cmd UNUSED)
+{
+ const int hsize=11;
+ const int vsize=cmd->n_variables*2+3;
+
+ assert(self);
+ self->populate = trbox_independent_samples_populate;
+
+ trbox_base_init(self,cmd->n_variables*2,hsize);
+ tab_title(self->t,0,_("Independent Samples Test"));
+ tab_hline(self->t,TAL_1,2,hsize-1,1);
+ tab_vline(self->t,TAL_2,2,0,vsize-1);
+ tab_vline(self->t,TAL_1,4,0,vsize-1);
+ tab_box(self->t,-1,-1,-1,TAL_1, 2,1,hsize-2,vsize-1);
+ tab_hline(self->t,TAL_1, hsize-2,hsize-1,2);
+ tab_box(self->t,-1,-1,-1,TAL_1, hsize-2,2,hsize-1,vsize-1);
+ tab_joint_text(self->t, 2, 0, 3, 0,
+ TAB_CENTER,_("Levene's Test for Equality of Variances"));
+ tab_joint_text(self->t, 4,0,hsize-1,0,
+ TAB_CENTER,_("t-test for Equality of Means"));
+
+ tab_text(self->t,2,2, TAB_CENTER | TAT_TITLE,_("F"));
+ tab_text(self->t,3,2, TAB_CENTER | TAT_TITLE,_("Sig."));
+ tab_text(self->t,4,2, TAB_CENTER | TAT_TITLE,_("t"));
+ tab_text(self->t,5,2, TAB_CENTER | TAT_TITLE,_("df"));
+ tab_text(self->t,6,2, TAB_CENTER | TAT_TITLE,_("Sig. (2-tailed)"));
+ tab_text(self->t,7,2, TAB_CENTER | TAT_TITLE,_("Mean Difference"));
+ tab_text(self->t,8,2, TAB_CENTER | TAT_TITLE,_("Std. Error Difference"));
+ tab_text(self->t,9,2, TAB_CENTER | TAT_TITLE,_("Lower"));
+ tab_text(self->t,10,2, TAB_CENTER | TAT_TITLE,_("Upper"));
+
+ tab_joint_text(self->t, 9, 1, 10, 1, TAB_CENTER | TAT_PRINTF,
+ _("%g%% Confidence Interval of the Difference"),
+ cmd->criteria*100.0);
+
+}
+
+/* Populate the independent samples trbox */
+void
+trbox_independent_samples_populate(struct trbox *self,
+ struct cmd_t_test *cmd )
+{
+ int i;
+
+ assert(self);
+ for (i=0; i < cmd->n_variables; ++i)
+ {
+ double p,q;
+
+ double t;
+ double df;
+
+ double df1, df2;
+
+ double pooled_variance;
+ double std_err_diff;
+ double mean_diff;
+
+ struct group_statistics *gs0 = &cmd->v_variables[i]->p.t_t.gs[0];
+ struct group_statistics *gs1 = &cmd->v_variables[i]->p.t_t.gs[1];
+
+ tab_text (self->t, 0, i*2+3, TAB_LEFT, cmd->v_variables[i]->name);
+
+ tab_text (self->t, 1, i*2+3, TAB_LEFT, _("Equal variances assumed"));
+
+
+ tab_float(self->t, 2, i*2+3, TAB_CENTER,
+ cmd->v_variables[i]->p.t_t.levene, 8,3);
+
+ /* Now work out the significance of the Levene test */
+ df1 = 1; df2 = cmd->v_variables[i]->p.t_t.ugs.n - 2;
+ q = gsl_cdf_fdist_Q(cmd->v_variables[i]->p.t_t.levene, df1, df2);
+
+ tab_float(self->t, 3, i*2+3, TAB_CENTER, q, 8,3 );
+
+ df = gs0->n + gs1->n - 2.0 ;
+ tab_float (self->t, 5, i*2+3, TAB_RIGHT, df, 2, 0);
+
+ pooled_variance = ( (gs0->n )*pow2(gs0->s_std_dev)
+ +
+ (gs1->n )*pow2(gs1->s_std_dev)
+ ) / df ;
+
+ t = (gs0->mean - gs1->mean) / sqrt(pooled_variance) ;
+ t /= sqrt((gs0->n + gs1->n)/(gs0->n*gs1->n));
+
+ tab_float (self->t, 4, i*2+3, TAB_RIGHT, t, 8, 3);
+
+ p = gsl_cdf_tdist_P(t, df);
+ q = gsl_cdf_tdist_Q(t, df);
+
+ tab_float(self->t, 6, i*2+3, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
+
+ mean_diff = gs0->mean - gs1->mean;
+ tab_float(self->t, 7, i*2+3, TAB_RIGHT, mean_diff, 8, 3);
+
+
+ std_err_diff = sqrt( pow2(gs0->se_mean) + pow2(gs1->se_mean));
+ tab_float(self->t, 8, i*2+3, TAB_RIGHT, std_err_diff, 8, 3);
+
+
+ /* Now work out the confidence interval */
+ q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
+
+ t = gsl_cdf_tdist_Qinv(q,df);
+ tab_float(self->t, 9, i*2+3, TAB_RIGHT,
+ mean_diff - t * std_err_diff, 8, 3);
+
+ tab_float(self->t, 10, i*2+3, TAB_RIGHT,
+ mean_diff + t * std_err_diff, 8, 3);
+
+
{
- printf ("P - Q NE 1\n");
- break;
+ double se2;
+ /* Now for the \sigma_1 != \sigma_2 case */
+ tab_text (self->t, 1, i*2+3+1,
+ TAB_LEFT, _("Equal variances not assumed"));
+
+
+ se2 = (pow2(gs0->s_std_dev)/(gs0->n -1) ) +
+ (pow2(gs1->s_std_dev)/(gs1->n -1) );
+
+ t = mean_diff / sqrt(se2) ;
+ tab_float (self->t, 4, i*2+3+1, TAB_RIGHT, t, 8, 3);
+
+ df = pow2(se2) / (
+ (pow2(pow2(gs0->s_std_dev)/(gs0->n - 1 ))
+ /(gs0->n -1 )
+ )
+ +
+ (pow2(pow2(gs1->s_std_dev)/(gs1->n - 1 ))
+ /(gs1->n -1 )
+ )
+ ) ;
+ tab_float (self->t, 5, i*2+3+1, TAB_RIGHT, df, 8, 3);
+
+ p = gsl_cdf_tdist_P(t, df);
+ q = gsl_cdf_tdist_Q(t, df);
+
+ tab_float(self->t, 6, i*2+3+1, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
+
+ /* Now work out the confidence interval */
+ q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
+
+ t = gsl_cdf_tdist_Qinv(q, df);
+
+ tab_float(self->t, 7, i*2+3+1, TAB_RIGHT, mean_diff, 8, 3);
+
+
+ tab_float(self->t, 8, i*2+3+1, TAB_RIGHT, std_err_diff, 8, 3);
+
+
+ tab_float(self->t, 9, i*2+3+1, TAB_RIGHT,
+ mean_diff - t * std_err_diff, 8, 3);
+
+ tab_float(self->t, 10, i*2+3+1, TAB_RIGHT,
+ mean_diff + t * std_err_diff, 8, 3);
+
}
}
+}
- if (status)
+/* Initialize the paired samples trbox */
+void
+trbox_paired_init(struct trbox *self,
+ struct cmd_t_test *cmd UNUSED)
+{
+
+ const int hsize=10;
+ const int vsize=n_pairs+3;
+
+ self->populate = trbox_paired_populate;
+
+ trbox_base_init(self,n_pairs,hsize);
+ tab_title (self->t, 0, _("Paired Samples Test"));
+ tab_hline(self->t,TAL_1,2,6,1);
+ tab_vline(self->t,TAL_2,2,0,vsize);
+ tab_joint_text(self->t,2,0,6,0,TAB_CENTER,_("Paired Differences"));
+ tab_box(self->t,-1,-1,-1,TAL_1, 2,1,6,vsize-1);
+ tab_box(self->t,-1,-1,-1,TAL_1, 6,0,hsize-1,vsize-1);
+ tab_hline(self->t,TAL_1,5,6, 2);
+ tab_vline(self->t,TAL_0,6,0,1);
+
+ tab_joint_text(self->t, 5, 1, 6, 1, TAB_CENTER | TAT_PRINTF,
+ _("%g%% Confidence Interval of the Difference"),
+ cmd->criteria*100.0);
+
+ tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _("Mean"));
+ tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
+ tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _("Std. Error Mean"));
+ tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _("Lower"));
+ tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _("Upper"));
+ tab_text (self->t, 7, 2, TAB_CENTER | TAT_TITLE, _("t"));
+ tab_text (self->t, 8, 2, TAB_CENTER | TAT_TITLE, _("df"));
+ tab_text (self->t, 9, 2, TAB_CENTER | TAT_TITLE, _("Sig. (2-tailed)"));
+}
+
+/* Populate the paired samples trbox */
+void
+trbox_paired_populate(struct trbox *trb,
+ struct cmd_t_test *cmd UNUSED)
+{
+ int i;
+
+ for (i=0; i < n_pairs; ++i)
{
- return (double) ERROR_SIG;
+ double p,q;
+ double se_mean;
+
+ double n = pairs[i].n;
+ double t;
+ double df = n - 1;
+
+ tab_text (trb->t, 0, i+3, TAB_LEFT | TAT_PRINTF, _("Pair %d"),i);
+
+ tab_text (trb->t, 1, i+3, TAB_LEFT | TAT_PRINTF, "%s - %s",
+ pairs[i].v[0]->name, pairs[i].v[1]->name);
+
+ tab_float(trb->t, 2, i+3, TAB_RIGHT, pairs[i].mean_diff, 8, 4);
+
+ tab_float(trb->t, 3, i+3, TAB_RIGHT, pairs[i].std_dev_diff, 8, 5);
+
+ /* SE Mean */
+ se_mean = pairs[i].std_dev_diff / sqrt(n) ;
+ tab_float(trb->t, 4, i+3, TAB_RIGHT, se_mean, 8,5 );
+
+ /* Now work out the confidence interval */
+ q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
+
+ t = gsl_cdf_tdist_Qinv(q, df);
+
+ tab_float(trb->t, 5, i+3, TAB_RIGHT,
+ pairs[i].mean_diff - t * se_mean , 8, 4);
+
+ tab_float(trb->t, 6, i+3, TAB_RIGHT,
+ pairs[i].mean_diff + t * se_mean , 8, 4);
+
+ t = (pairs[i].mean[0] - pairs[i].mean[1])
+ / sqrt (
+ ( pow2 (pairs[i].s_std_dev[0]) + pow2 (pairs[i].s_std_dev[1]) -
+ 2 * pairs[i].correlation *
+ pairs[i].s_std_dev[0] * pairs[i].s_std_dev[1] )
+ / (n - 1)
+ );
+
+ tab_float(trb->t, 7, i+3, TAB_RIGHT, t , 8,3 );
+
+ /* Degrees of freedom */
+ tab_float(trb->t, 8, i+3, TAB_RIGHT, df , 2, 0 );
+
+ p = gsl_cdf_tdist_P(t,df);
+ q = gsl_cdf_tdist_P(t,df);
+
+ tab_float(trb->t, 9, i+3, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
+
}
- else
+}
+
+/* Initialize the one sample trbox */
+void
+trbox_one_sample_init(struct trbox *self, struct cmd_t_test *cmd )
+{
+ const int hsize=7;
+ const int vsize=cmd->n_variables+3;
+
+ self->populate = trbox_one_sample_populate;
+
+ trbox_base_init(self, cmd->n_variables,hsize);
+ tab_title (self->t, 0, _("One-Sample Test"));
+ tab_hline(self->t, TAL_1, 1, hsize - 1, 1);
+ tab_vline(self->t, TAL_2, 1, 0, vsize);
+
+ tab_joint_text(self->t, 1, 0, hsize-1,0, TAB_CENTER | TAT_PRINTF,
+ _("Test Value = %f"),cmd->n_testval);
+
+ tab_box(self->t, -1, -1, -1, TAL_1, 1,1,hsize-1,vsize-1);
+
+
+ tab_joint_text(self->t,5,1,6,1,TAB_CENTER | TAT_PRINTF,
+ _("%g%% Confidence Interval of the Difference"),
+ cmd->criteria*100.0);
+
+ tab_vline(self->t,TAL_0,6,1,1);
+ tab_hline(self->t,TAL_1,5,6,2);
+ tab_text (self->t, 1, 2, TAB_CENTER | TAT_TITLE, _("t"));
+ tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _("df"));
+ tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _("Sig. (2-tailed)"));
+ tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _("Mean Difference"));
+ tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _("Lower"));
+ tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _("Upper"));
+
+}
+
+
+/* Populate the one sample trbox */
+void
+trbox_one_sample_populate(struct trbox *trb, struct cmd_t_test *cmd)
+{
+ int i;
+
+ assert(trb->t);
+
+ for (i=0; i < cmd->n_variables; ++i)
{
- return q;
+ double t;
+ double p,q;
+ double df;
+ struct group_statistics *gs;
+ gs= &cmd->v_variables[i]->p.t_t.ugs;
+
+
+ tab_text (trb->t, 0, i+3, TAB_LEFT, cmd->v_variables[i]->name);
+
+ t = (gs->mean - cmd->n_testval ) * sqrt(gs->n) / gs->std_dev ;
+
+ tab_float (trb->t, 1, i+3, TAB_RIGHT, t, 8,3);
+
+ /* degrees of freedom */
+ df = gs->n - 1;
+
+ tab_float (trb->t, 2, i+3, TAB_RIGHT, df, 8,0);
+
+ p = gsl_cdf_tdist_P(t, df);
+ q = gsl_cdf_tdist_Q(t, df);
+
+ /* Multiply by 2 to get 2-tailed significance, makeing sure we've got
+ the correct tail*/
+ tab_float (trb->t, 3, i+3, TAB_RIGHT, 2.0*(t>0?q:p), 8,3);
+
+ tab_float (trb->t, 4, i+3, TAB_RIGHT, gs->mean_diff, 8,3);
+
+
+ q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
+ t = gsl_cdf_tdist_Qinv(q, df);
+
+ tab_float (trb->t, 5, i+3, TAB_RIGHT,
+ gs->mean_diff - t * gs->se_mean, 8,4);
+
+ tab_float (trb->t, 6, i+3, TAB_RIGHT,
+ gs->mean_diff + t * gs->se_mean, 8,4);
}
}
-double
-t_crt (double df, double q)
+/* Base initializer for the generalized trbox */
+void
+trbox_base_init(struct trbox *self, int data_rows, int cols)
{
- int which, status;
- double p, bound, t;
+ const int rows = 3 + data_rows;
+
+ self->finalize = trbox_base_finalize;
+ self->t = tab_create (cols, rows, 0);
+ tab_headers (self->t,0,0,3,0);
+ tab_box (self->t, TAL_2, TAL_2, TAL_0, TAL_0, 0, 0, cols -1, rows -1);
+ tab_hline(self->t, TAL_2,0,cols-1,3);
+ tab_dim (self->t, tab_natural_dimensions);
+}
- which = FIND_CRITICAL_VALUE;
- bound = 0.0;
- p = 1.0 - q;
- t = 0.0;
- cdft (&which, &p, &q, &t, &df, &status, &bound);
+/* Base finalizer for the trbox */
+void
+trbox_base_finalize(struct trbox *trb)
+{
+ tab_submit(trb->t);
+}
- switch (status)
+
+/* Create , populate and submit the Paired Samples Correlation box */
+void
+pscbox(void)
+{
+ const int rows=1+n_pairs;
+ const int cols=5;
+ int i;
+
+ struct tab_table *table;
+
+ table = tab_create (cols,rows,0);
+
+ tab_columns (table, SOM_COL_DOWN, 1);
+ tab_headers (table,0,0,1,0);
+ tab_box (table, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols -1, rows -1 );
+ tab_hline(table, TAL_2, 0, cols - 1, 1);
+ tab_vline(table, TAL_2, 2, 0, rows - 1);
+ tab_dim(table, tab_natural_dimensions);
+ tab_title(table, 0, _("Paired Samples Correlations"));
+
+ /* column headings */
+ tab_text(table, 2,0, TAB_CENTER | TAT_TITLE, _("N"));
+ tab_text(table, 3,0, TAB_CENTER | TAT_TITLE, _("Correlation"));
+ tab_text(table, 4,0, TAB_CENTER | TAT_TITLE, _("Sig."));
+
+ for (i=0; i < n_pairs; ++i)
{
- case -1:
- {
- printf ("t_crt: Parameter 1 is out of range\n");
- break;
- }
- case -2:
- {
- printf ("t_crt: value of p (%f) is out of range\n", p);
- break;
- }
- case -3:
- {
- printf ("t_crt: value of q (%f) is out of range\n", q);
- break;
- }
- case -4:
- {
- printf ("t_crt: value of df (%f) is out of range\n", df);
- break;
- }
- case -5:
- {
- printf ("t_crt: Parameter 5 is out of range\n");
- break;
- }
- case -6:
- {
- printf ("t_crt: Parameter 6 is out of range\n");
- break;
- }
- case -7:
- {
- printf ("t_crt: Parameter 7 is out of range\n");
- break;
- }
- case 0:
- {
- /* printf( "Command completed successfully\n" ); */
- break;
- }
- case 1:
- {
- printf ("t_crt: Answer appears to be lower than the lowest search bound\n");
- break;
- }
- case 2:
- {
- printf ("t_crt: Answer appears to be higher than the greatest search bound\n");
- break;
- }
- case 3:
- {
- printf ("t_crt: P - Q NE 1\n");
- break;
- }
+ double p,q;
+
+ double df = pairs[i].n -2;
+
+ double correlation_t =
+ pairs[i].correlation * sqrt(df) /
+ sqrt(1 - pow2(pairs[i].correlation));
+
+
+ /* row headings */
+ tab_text(table, 0,i+1, TAB_LEFT | TAT_TITLE | TAT_PRINTF,
+ _("Pair %d"), i);
+
+ tab_text(table, 1,i+1, TAB_LEFT | TAT_TITLE | TAT_PRINTF,
+ _("%s & %s"), pairs[i].v[0]->name, pairs[i].v[1]->name);
+
+
+ /* row data */
+ tab_float(table, 2, i+1, TAB_RIGHT, pairs[i].n, 4, 0);
+ tab_float(table, 3, i+1, TAB_RIGHT, pairs[i].correlation, 8, 3);
+
+ p = gsl_cdf_tdist_P(correlation_t, df);
+ q = gsl_cdf_tdist_Q(correlation_t, df);
+
+ tab_float(table, 4, i+1, TAB_RIGHT, 2.0*(correlation_t>0?q:p), 8, 3);
}
- if (status)
+ tab_submit(table);
+}
+
+
+
+/* Calculation Implementation */
+
+/* Per case calculations common to all variants of the T test */
+static int
+common_calc (const struct ccase *c, void *_cmd)
+{
+ int i;
+ struct cmd_t_test *cmd = (struct cmd_t_test *)_cmd;
+
+ double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
+
+
+ /* Skip the entire case if /MISSING=LISTWISE is set */
+ if ( cmd->miss == TTS_LISTWISE )
{
- return (double) ERROR_SIG;
+ for(i=0; i< cmd->n_variables ; ++i)
+ {
+ struct variable *v = cmd->v_variables[i];
+ const union value *val = &c->data[v->fv];
+
+ if (value_is_missing(val,v) )
+ {
+ return 0;
+ }
+ }
}
- else
+
+ /* Listwise has to be implicit if the independent variable is missing ?? */
+ if ( cmd->sbc_groups )
{
- return t;
+ const union value *gv = &c->data[indep_var->fv];
+ if ( value_is_missing(gv,indep_var) )
+ {
+ return 0;
+ }
}
+
+
+ for(i=0; i< cmd->n_variables ; ++i)
+ {
+ struct group_statistics *gs;
+ struct variable *v = cmd->v_variables[i];
+ const union value *val = &c->data[v->fv];
+
+ gs= &cmd->v_variables[i]->p.t_t.ugs;
+
+ if (! value_is_missing(val,v) )
+ {
+ gs->n+=weight;
+ gs->sum+=weight * val->f;
+ gs->ssq+=weight * val->f * val->f;
+ }
+ }
+ return 0;
}
-double
-t_sig (double t, double df)
+/* Pre calculations common to all variants of the T test */
+static void
+common_precalc ( struct cmd_t_test *cmd )
{
- int which, status;
- double p, q, bound;
+ int i=0;
+
+ for(i=0; i< cmd->n_variables ; ++i)
+ {
+ struct group_statistics *gs;
+ gs= &cmd->v_variables[i]->p.t_t.ugs;
+
+ gs->sum=0;
+ gs->n=0;
+ gs->ssq=0;
+ gs->sum_diff=0;
+ }
+}
- which = FIND_P;
- q = 0.0;
- p = 0.0;
- bound = 0.0;
+/* Post calculations common to all variants of the T test */
+void
+common_postcalc ( struct cmd_t_test *cmd )
+{
+ int i=0;
- cdft (&which, &p, &q, &t, &df, &status, &bound);
- switch (status)
+ for(i=0; i< cmd->n_variables ; ++i)
{
- case -1:
- {
- printf ("t-sig: Parameter 1 is out of range\n");
- break;
- }
- case -2:
- {
- printf ("t-sig: Parameter 2 is out of range\n");
- break;
- }
- case -3:
- {
- printf ("t-sig: Parameter 3 is out of range\n");
- break;
- }
- case -4:
- {
- printf ("t-sig: Parameter 4 is out of range\n");
- break;
- }
- case -5:
- {
- printf ("t-sig: Parameter 5 is out of range\n");
- break;
- }
- case -6:
- {
- printf ("t-sig: Parameter 6 is out of range\n");
- break;
- }
- case -7:
- {
- printf ("t-sig: Parameter 7 is out of range\n");
- break;
- }
- case 0:
- {
- /* printf( "Command completed successfully\n" ); */
- break;
- }
- case 1:
- {
- printf ("t-sig: Answer appears to be lower than the lowest search bound\n");
- break;
- }
- case 2:
- {
- printf ("t-sig: Answer appears to be higher than the greatest search bound\n");
- break;
- }
- case 3:
+ struct group_statistics *gs;
+ gs= &cmd->v_variables[i]->p.t_t.ugs;
+
+ gs->mean=gs->sum / gs->n;
+ gs->s_std_dev= sqrt(
+ ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
+ ) ;
+
+ gs->std_dev= sqrt(
+ gs->n/(gs->n-1) *
+ ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
+ ) ;
+
+ gs->se_mean = gs->std_dev / sqrt(gs->n);
+ gs->mean_diff= gs->sum_diff / gs->n;
+ }
+}
+
+/* Per case calculations for one sample t test */
+static int
+one_sample_calc (const struct ccase *c, void *cmd_)
+{
+ int i;
+ struct cmd_t_test *cmd = (struct cmd_t_test *)cmd_;
+
+
+ double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
+
+ /* Skip the entire case if /MISSING=LISTWISE is set */
+ if ( cmd->miss == TTS_LISTWISE )
+ {
+ for(i=0; i< cmd->n_variables ; ++i)
+ {
+ struct variable *v = cmd->v_variables[i];
+ const union value *val = &c->data[v->fv];
+
+ if (value_is_missing(val,v) )
+ {
+ return 0;
+ }
+ }
+ }
+
+ for(i=0; i< cmd->n_variables ; ++i)
+ {
+ struct group_statistics *gs;
+ struct variable *v = cmd->v_variables[i];
+ const union value *val = &c->data[v->fv];
+
+ gs= &cmd->v_variables[i]->p.t_t.ugs;
+
+ if ( ! value_is_missing(val,v))
+ gs->sum_diff += weight * (val->f - cmd->n_testval);
+ }
+
+ return 0;
+}
+
+/* Pre calculations for one sample t test */
+static void
+one_sample_precalc ( struct cmd_t_test *cmd )
+{
+ int i=0;
+
+ for(i=0; i< cmd->n_variables ; ++i)
+ {
+ struct group_statistics *gs;
+ gs= &cmd->v_variables[i]->p.t_t.ugs;
+
+ gs->sum_diff=0;
+ }
+}
+
+/* Post calculations for one sample t test */
+static void
+one_sample_postcalc (struct cmd_t_test *cmd)
+{
+ int i=0;
+
+ for(i=0; i< cmd->n_variables ; ++i)
+ {
+ struct group_statistics *gs;
+ gs= &cmd->v_variables[i]->p.t_t.ugs;
+
+ gs->mean_diff = gs->sum_diff / gs->n ;
+ }
+}
+
+
+
+static int
+compare_var_name (const void *a_, const void *b_, void *v_ UNUSED)
+{
+ const struct variable *a = a_;
+ const struct variable *b = b_;
+
+ return strcmp(a->name,b->name);
+}
+
+static unsigned
+hash_var_name (const void *a_, void *v_ UNUSED)
+{
+ const struct variable *a = a_;
+
+ return hsh_hash_bytes (a->name, strlen(a->name));
+}
+
+
+
+static void
+paired_precalc (struct cmd_t_test *cmd UNUSED)
+{
+ int i;
+
+ for(i=0; i < n_pairs ; ++i )
+ {
+ pairs[i].n = 0;
+ pairs[i].sum[0] = 0; pairs[i].sum[1] = 0;
+ pairs[i].ssq[0] = 0; pairs[i].ssq[1] = 0;
+ pairs[i].sum_of_prod = 0;
+ pairs[i].correlation = 0;
+ pairs[i].sum_of_diffs = 0;
+ pairs[i].ssq_diffs = 0;
+ }
+
+}
+
+
+static int
+paired_calc (const struct ccase *c, void *cmd_)
+{
+ int i;
+
+ struct cmd_t_test *cmd = (struct cmd_t_test *) cmd_;
+
+ double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
+
+ /* Skip the entire case if /MISSING=LISTWISE is set ,
+ AND one member of a pair is missing */
+ if ( cmd->miss == TTS_LISTWISE )
+ {
+ for(i=0; i < n_pairs ; ++i )
+ {
+ struct variable *v0 = pairs[i].v[0];
+ struct variable *v1 = pairs[i].v[1];
+
+ const union value *val0 = &c->data[v0->fv];
+ const union value *val1 = &c->data[v1->fv];
+
+ if ( value_is_missing(val0,v0) ||
+ value_is_missing(val1,v1) )
+ {
+ return 0;
+ }
+ }
+ }
+
+ for(i=0; i < n_pairs ; ++i )
+ {
+ struct variable *v0 = pairs[i].v[0];
+ struct variable *v1 = pairs[i].v[1];
+
+ const union value *val0 = &c->data[v0->fv];
+ const union value *val1 = &c->data[v1->fv];
+
+ if ( ( !value_is_missing(val0,v0) && !value_is_missing(val1,v1) ) )
{
- printf ("t-sig: P - Q NE 1\n");
- break;
+ pairs[i].n += weight;
+ pairs[i].sum[0] += weight * val0->f;
+ pairs[i].sum[1] += weight * val1->f;
+
+ pairs[i].ssq[0] += weight * pow2(val0->f);
+ pairs[i].ssq[1] += weight * pow2(val1->f);
+
+ pairs[i].sum_of_prod += weight * val0->f * val1->f ;
+
+ pairs[i].sum_of_diffs += weight * ( val0->f - val1->f ) ;
+ pairs[i].ssq_diffs += weight * pow2(val0->f - val1->f);
}
}
- if (status)
+ return 0;
+}
+
+static void
+paired_postcalc (struct cmd_t_test *cmd UNUSED)
+{
+ int i;
+
+ for(i=0; i < n_pairs ; ++i )
{
- return (double) ERROR_SIG;
+ int j;
+ const double n = pairs[i].n;
+
+ for (j=0; j < 2 ; ++j)
+ {
+ pairs[i].mean[j] = pairs[i].sum[j] / n ;
+ pairs[i].s_std_dev[j] = sqrt((pairs[i].ssq[j] / n -
+ pow2(pairs[i].mean[j]))
+ );
+
+ pairs[i].std_dev[j] = sqrt(n/(n-1)*(pairs[i].ssq[j] / n -
+ pow2(pairs[i].mean[j]))
+ );
+ }
+
+ pairs[i].correlation = pairs[i].sum_of_prod / pairs[i].n -
+ pairs[i].mean[0] * pairs[i].mean[1] ;
+ /* correlation now actually contains the covariance */
+
+ pairs[i].correlation /= pairs[i].std_dev[0] * pairs[i].std_dev[1];
+ pairs[i].correlation *= pairs[i].n / ( pairs[i].n - 1 );
+
+ pairs[i].mean_diff = pairs[i].sum_of_diffs / n ;
+
+ pairs[i].std_dev_diff = sqrt ( n / (n - 1) * (
+ ( pairs[i].ssq_diffs / n )
+ -
+ pow2(pairs[i].mean_diff )
+ ) );
}
- else
+}
+
+/* Return the group # corresponding to the
+ independent variable with the value val
+*/
+static int
+get_group(const union value *val, struct variable *indep)
+{
+ int i;
+
+ for (i = 0; i < 2 ; ++i )
{
- return q;
+ const int cmp = compare_values(val,&groups_values[i],indep->width) ;
+ switch ( criteria[i])
+ {
+ case CMP_EQ:
+ if ( 0 == cmp ) return i;
+ break;
+ case CMP_LT:
+ if ( 0 > cmp ) return i;
+ break;
+ case CMP_LE:
+ if ( cmp <= 0 ) return i;
+ break;
+ case CMP_GT:
+ if ( cmp > 0 ) return i;
+ break;
+ case CMP_GE:
+ if ( cmp >= 0 ) return i;
+ break;
+ default:
+ assert(0);
+ };
}
+
+ /* No groups matched */
+ return -1;
}
-double
-covariance (double x_sum, double x_n, double y_sum, double y_n, double ss)
+
+static void
+group_precalc (struct cmd_t_test *cmd )
{
- double tmp;
+ int i;
+ int j;
+
+ for(i=0; i< cmd->n_variables ; ++i)
+ {
+ struct t_test_proc *ttpr = &cmd->v_variables[i]->p.t_t;
+
+ /* There's always 2 groups for a T - TEST */
+ ttpr->n_groups = 2;
+ ttpr->gs = xmalloc(sizeof(struct group_statistics) * 2) ;
+
+ for (j=0 ; j < 2 ; ++j)
+ {
+ ttpr->gs[j].sum = 0;
+ ttpr->gs[j].n = 0;
+ ttpr->gs[j].ssq = 0;
+
+ if ( n_group_values == 2 )
+ ttpr->gs[j].id = groups_values[j];
+ else
+ ttpr->gs[j].id = groups_values[0];
+ ttpr->gs[j].criterion = criteria[j];
+ }
+ }
- tmp = x_sum * y_sum;
- tmp = tmp / x_n;
- tmp = ss - tmp;
- tmp = (tmp / (x_n + y_n - 1.0));
- return tmp;
}
-double
-pearson_r (double c_xy, double c_xx, double c_yy)
+static int
+group_calc (const struct ccase *c, struct cmd_t_test *cmd)
{
- return (c_xy / (sqrt (c_xx * c_yy)));
+ int i;
+ int g;
+
+ const union value *gv = &c->data[indep_var->fv];
+
+ const double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
+
+ if ( value_is_missing(gv,indep_var) )
+ {
+ return 0;
+ }
+
+ if ( cmd->miss == TTS_LISTWISE )
+ {
+ for(i=0; i< cmd->n_variables ; ++i)
+ {
+ struct variable *v = cmd->v_variables[i];
+ const union value *val = &c->data[v->fv];
+
+ if (value_is_missing(val,v) )
+ {
+ return 0;
+ }
+ }
+ }
+
+
+ gv = &c->data[indep_var->fv];
+
+ g = get_group(gv,indep_var);
+
+
+ /* If the independent variable doesn't match either of the values
+ for this case then move on to the next case */
+ if (g == -1 )
+ return 0;
+
+ for(i=0; i< cmd->n_variables ; ++i)
+ {
+ struct variable *var = cmd->v_variables[i];
+
+ struct group_statistics *gs = &var->p.t_t.gs[g];
+
+ const union value *val=&c->data[var->fv];
+
+ if ( !value_is_missing(val,var) )
+ {
+ gs->n+=weight;
+ gs->sum+=weight * val->f;
+ gs->ssq+=weight * pow2(val->f);
+ }
+ }
+
+ return 0;
}
-void
-print_t_groups (struct variable * grps, union value * g1, union value * g2,
- double n1, double n2, double mean1, double mean2,
- double sd1, double sd2, double se1, double se2,
- double diff, double l_f, double l_p,
- double p_t, double p_sig, double p_df, double p_sed,
- double p_l, double p_h,
- double s_t, double s_sig, double s_df, double s_sed,
- double s_l, double s_h)
+
+static void
+group_postcalc ( struct cmd_t_test *cmd )
{
+ int i;
+ int j;
- /* Display all this shit as SPSS 6.0 does (roughly) */
- printf ("\n\n Number \n");
- printf (" Variable of Cases Mean SD SE of Mean\n");
- printf ("-----------------------------------------------------------\n");
- printf (" %s %s\n\n", cmd.v_variables[cur_var]->name, cmd.v_variables[cur_var]->label);
- printf ("%s %8.4f %8.0f %8.4f %8.3f %8.3f\n",
- get_val_lab (grps, *g1, 0), g1->f, n1, mean1, sd1, se1);
- printf ("%s %8.4f %8.0f %8.4f %8.3f %8.3f\n",
- get_val_lab (grps, *g2, 0), g2->f, n2, mean2, sd2, se2);
- printf ("-----------------------------------------------------------\n");
- printf ("\n Mean Difference = %8.4f\n", diff);
- printf ("\n Levene's Test for Equality of Variances: F= %.3f P= %.3f\n",
- l_f, l_p);
- printf ("\n\n t-test for Equality of Means 95pc \n");
- printf ("Variances t-value df 2-Tail Sig SE of Diff CI for Diff \n");
- printf ("-----------------------------------------------------------------\n");
- printf ("Equal %8.2f %8.0f %8.3f %8.3f (%8.3f, %8.3f)\n",
- p_t, p_df, p_sig, p_sed, p_l, p_h);
- printf ("Unequal %8.2f %8.2f %8.3f %8.3f (%8.3f, %8.3f)\n",
- s_t, s_df, s_sig, s_sed, s_l, s_h);
- printf ("-----------------------------------------------------------------\n");
+ for(i=0; i< cmd->n_variables ; ++i)
+ {
+ for (j=0 ; j < 2 ; ++j)
+ {
+ struct group_statistics *gs;
+ gs=&cmd->v_variables[i]->p.t_t.gs[j];
+
+ gs->mean = gs->sum / gs->n;
+
+ gs->s_std_dev= sqrt(
+ ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
+ ) ;
+
+ gs->std_dev= sqrt(
+ gs->n/(gs->n-1) *
+ ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
+ ) ;
+
+ gs->se_mean = gs->std_dev / sqrt(gs->n);
+ }
+ }
}
-/*
- Local Variables:
- mode: c
- End:
-*/
+
+
+static void
+calculate(const struct casefile *cf, void *cmd_)
+{
+ struct ssbox stat_summary_box;
+ struct trbox test_results_box;
+
+ struct casereader *r;
+ const struct ccase *c;
+
+ struct cmd_t_test *cmd = (struct cmd_t_test *) cmd_;
+
+ common_precalc(cmd);
+ for(r = casefile_get_reader (cf);
+ casereader_read (r, &c) ; )
+ {
+ common_calc(c,cmd);
+ }
+ casereader_destroy (r);
+ common_postcalc(cmd);
+
+ switch(mode)
+ {
+ case T_1_SAMPLE:
+ one_sample_precalc(cmd);
+ for(r = casefile_get_reader (cf);
+ casereader_read (r, &c) ; )
+ {
+ one_sample_calc(c,cmd);
+ }
+ casereader_destroy (r);
+ one_sample_postcalc(cmd);
+
+ break;
+ case T_PAIRED:
+ paired_precalc(cmd);
+ for(r = casefile_get_reader (cf);
+ casereader_read (r, &c) ; )
+ {
+ paired_calc(c,cmd);
+ }
+ casereader_destroy (r);
+ paired_postcalc(cmd);
+
+ break;
+ case T_IND_SAMPLES:
+
+ group_precalc(cmd);
+ for(r = casefile_get_reader (cf);
+ casereader_read (r, &c) ; )
+ {
+ group_calc(c,cmd);
+ }
+ casereader_destroy (r);
+ group_postcalc(cmd);
+
+
+ levene(cf, indep_var, cmd->n_variables, cmd->v_variables,
+ (cmd->miss == TTS_LISTWISE)?LEV_LISTWISE:LEV_ANALYSIS ,
+ value_is_missing);
+ break;
+ }
+
+ ssbox_create(&stat_summary_box,cmd,mode);
+ ssbox_populate(&stat_summary_box,cmd);
+ ssbox_finalize(&stat_summary_box);
+
+ if ( mode == T_PAIRED)
+ pscbox();
+
+ trbox_create(&test_results_box,cmd,mode);
+ trbox_populate(&test_results_box,cmd);
+ trbox_finalize(&test_results_box);
+
+}
projects / pspp-builds.git / blobdiff