@table @asis
@item Explicit categories.
-@anchor{CTABLE Explicit Category List}
+@anchor{CTABLES Explicit Category List}
To explicitly specify categories to include, list the categories
within square brackets in the desired sort order. Use spaces or
commas to separate values. Categories not covered by the list are
With @code{SMISSING=VARIABLE}, which is the default, missing values
are excluded on a variable-by-variable basis. With
-@code{SMISSING=LISTWISE}, when scalar variables are stacked, a missing
-value for any of the scalar variables causes the case to be excluded
-for all of them.
+@code{SMISSING=LISTWISE}, when stacked scalar variables are nested
+together with a categorical variable, a missing value for any of the
+scalar variables causes the case to be excluded for all of them.
+
+As an example, consider the following dataset, in which @samp{x} is a
+categorical variable and @samp{y} and @samp{z} are scale:
+
+@psppoutput{ctables18}
+
+@noindent
+With the default missing-value treatment, @samp{x}'s mean is 20, based
+on the values 10, 20, and 30, and @samp{y}'s mean is 50, based on 40,
+50, and 60:
+
+@example
+CTABLES /TABLE (y + z) > x.
+@end example
+@psppoutput{ctables19}
+
+@noindent
+By adding @code{SMISSING=LISTWISE}, only cases where @samp{y} and
+@samp{z} are both non-missing are considered, so @samp{x}'s mean
+becomes 15, as the average of 10 and 20, and @samp{y}'s mean becomes
+55, the average of 50 and 60:
+
+@example
+CTABLES /SMISSING LISTWISE /TABLE (y + z) > x.
+@end example
+@psppoutput{ctables20}
+
+@noindent
+Even with @code{SMISSING=LISTWISE}, if @samp{y} and @samp{z} are
+separately nested with @samp{x}, instead of using a single @samp{>}
+operator, missing values revert to being considered on a
+variable-by-variable basis:
+
+@example
+CTABLES /SMISSING LISTWISE /TABLE (y > x) + (z > x).
+@end example
+@psppoutput{ctables21}
@node CTABLES Computed Categories
@subsection Computed Categories
categories created using arithmetic on categories obtained from the
data. The @code{PCOMPUTE} subcommand defines computed categories,
which can then be used in two places: on @code{CATEGORIES} within an
-explicit category list (@pxref{CTABLE Explicit Category List}), and on
+explicit category list (@pxref{CTABLES Explicit Category List}), and on
the @code{PPROPERTIES} subcommand to define further properties for a
given postcompute.
@itemx MISSING
@itemx OTHERNM
These forms evaluate to the summary statistics for categories matching
-the given syntax, as described in previous sections (@pxref{CTABLE
+the given syntax, as described in previous sections (@pxref{CTABLES
Explicit Category List}). If more than one category matches, their
values are summed.
(VALIDN and TOTALN act differently for summarizing scale and categorical
variables.) */
bool is_scale;
+
+ /* If any of these optional additional scale variables are missing, then
+ treat 'var' as if it's missing too. This is for implementing
+ SMISSING=LISTWISE. */
+ struct variable **listwise_vars;
+ size_t n_listwise_vars;
};
static void ctables_summary_spec_set_clone (struct ctables_summary_spec_set *,
size_t scale_idx;
size_t *domains[N_CTDTS];
size_t n_domains[N_CTDTS];
+ size_t group_head;
struct ctables_summary_spec_set specs[N_CSVS];
};
for (size_t i = 0; i < s0.n; i++)
stack.nests[stack.n++] = s0.nests[i];
for (size_t i = 0; i < s1.n; i++)
- stack.nests[stack.n++] = s1.nests[i];
+ {
+ stack.nests[stack.n] = s1.nests[i];
+ stack.nests[stack.n].group_head += s0.n;
+ stack.n++;
+ }
assert (stack.n == s0.n + s1.n);
free (s0.nests);
free (s1.nests);
return stack;
}
+static struct ctables_stack
+var_fts (const struct ctables_axis *a)
+{
+ assert (!a->var.is_mrset);
+
+ struct variable **vars = xmalloc (sizeof *vars);
+ *vars = a->var.var;
+
+ struct ctables_nest *nest = xmalloc (sizeof *nest);
+ *nest = (struct ctables_nest) {
+ .vars = vars,
+ .n = 1,
+ .scale_idx = a->scale ? 0 : SIZE_MAX,
+ };
+ if (a->specs[CSV_CELL].n || a->scale)
+ for (enum ctables_summary_variant sv = 0; sv < N_CSVS; sv++)
+ {
+ ctables_summary_spec_set_clone (&nest->specs[sv], &a->specs[sv]);
+ nest->specs[sv].var = a->var.var;
+ nest->specs[sv].is_scale = a->scale;
+ }
+ return (struct ctables_stack) { .nests = nest, .n = 1 };
+}
+
static struct ctables_stack
enumerate_fts (enum pivot_axis_type axis_type, const struct ctables_axis *a)
{
switch (a->op)
{
case CTAO_VAR:
- assert (!a->var.is_mrset);
-
- struct variable **vars = xmalloc (sizeof *vars);
- *vars = a->var.var;
-
- struct ctables_nest *nest = xmalloc (sizeof *nest);
- *nest = (struct ctables_nest) {
- .vars = vars,
- .n = 1,
- .scale_idx = a->scale ? 0 : SIZE_MAX,
- };
- if (a->specs[CSV_CELL].n || a->scale)
- for (enum ctables_summary_variant sv = 0; sv < N_CSVS; sv++)
- {
- ctables_summary_spec_set_clone (&nest->specs[sv], &a->specs[sv]);
- nest->specs[sv].var = a->var.var;
- nest->specs[sv].is_scale = a->scale;
- }
- return (struct ctables_stack) { .nests = nest, .n = 1 };
+ return var_fts (a);
case CTAO_STACK:
return stack_fts (enumerate_fts (axis_type, a->subs[0]),
enumerate_fts (axis_type, a->subs[1]));
case CTAO_NEST:
+ /* This should consider any of the scale variables found in the result to
+ be linked to each other listwise for SMISSING=LISTWISE. */
return nest_fts (enumerate_fts (axis_type, a->subs[0]),
enumerate_fts (axis_type, a->subs[1]));
}
ctables_summary_add (union ctables_summary *s,
const struct ctables_summary_spec *ss,
const struct variable *var, const union value *value,
- bool is_scale, bool is_missing, bool excluded_missing,
+ bool is_scale, bool is_scale_missing,
+ bool is_missing, bool excluded_missing,
double d_weight, double e_weight)
{
/* To determine whether a case is included in a given table for a particular
case CTSF_LAYERROWPCT_VALIDN:
case CTSF_LAYERCOLPCT_VALIDN:
if (is_scale
- ? !var_is_value_missing (var, value)
+ ? !is_scale_missing
: !is_missing)
s->count += d_weight;
break;
case CTSF_EVALIDN:
if (is_scale
- ? !var_is_value_missing (var, value)
+ ? !is_scale_missing
: !is_missing)
s->count += e_weight;
break;
case CTSF_MAXIMUM:
case CTSF_MINIMUM:
case CTSF_RANGE:
- if (!var_is_value_missing (var, value))
+ if (!is_scale_missing)
{
assert (!var_is_alpha (var)); /* XXX? */
if (s->min == SYSMIS || value->f < s->min)
case CTSF_LAYERPCT_SUM:
case CTSF_LAYERROWPCT_SUM:
case CTSF_LAYERCOLPCT_SUM:
- if (!var_is_value_missing (var, value))
+ if (!is_scale_missing)
moments1_add (s->moments, value->f, e_weight);
break;
case CTSF_MEDIAN:
case CTSF_MODE:
case CTSF_PTILE:
- if (var_is_value_missing (var, value))
+ if (!is_scale_missing)
{
s->ovalid += e_weight;
return cell;
}
+static bool
+is_scale_missing (const struct ctables_summary_spec_set *specs,
+ const struct ccase *c)
+{
+ if (!specs->is_scale)
+ return false;
+
+ if (var_is_num_missing (specs->var, case_num (c, specs->var)))
+ return true;
+
+ for (size_t i = 0; i < specs->n_listwise_vars; i++)
+ {
+ const struct variable *var = specs->listwise_vars[i];
+ if (var_is_num_missing (var, case_num (c, var)))
+ return true;
+ }
+
+ return false;
+}
+
static void
ctables_cell_add__ (struct ctables_section *s, const struct ccase *c,
const struct ctables_category *cats[PIVOT_N_AXES][10],
const struct ctables_nest *ss = s->nests[s->table->summary_axis];
const struct ctables_summary_spec_set *specs = &ss->specs[cell->sv];
+
+ bool scale_missing = is_scale_missing (specs, c);
for (size_t i = 0; i < specs->n; i++)
ctables_summary_add (&cell->summaries[i], &specs->specs[i],
specs->var, case_data (c, specs->var), specs->is_scale,
- is_missing, excluded_missing, d_weight, e_weight);
+ scale_missing, is_missing, excluded_missing,
+ d_weight, e_weight);
for (enum ctables_domain_type dt = 0; dt < N_CTDTS; dt++)
if (!(cell->omit_domains && (1u << dt)))
{
else if (!nest->specs[CSV_TOTAL].n)
ctables_summary_spec_set_clone (&nest->specs[CSV_TOTAL],
&nest->specs[CSV_CELL]);
+
+ if (t->ctables->smissing_listwise)
+ {
+ struct variable **listwise_vars = NULL;
+ size_t n = 0;
+ size_t allocated = 0;
+
+ for (size_t j = nest->group_head; j < stack->n; j++)
+ {
+ const struct ctables_nest *other_nest = &stack->nests[j];
+ if (other_nest->group_head != nest->group_head)
+ break;
+
+ if (nest != other_nest && other_nest->scale_idx < other_nest->n)
+ {
+ if (n >= allocated)
+ listwise_vars = x2nrealloc (listwise_vars, &allocated,
+ sizeof *listwise_vars);
+ listwise_vars[n++] = other_nest->vars[other_nest->scale_idx];
+ }
+ }
+ for (size_t j = 0; j < N_CSVS; j++)
+ {
+ nest->specs[j].listwise_vars = listwise_vars;
+ nest->specs[j].n_listwise_vars = n;
+ }
+ }
}
struct ctables_summary_spec_set *merged = &t->summary_specs;
projects / pspp / commitdiff