Census 1981: get an authoritative ruling from StatCan on the spouse-variable naming collision

Type: data quality / needs upstream confirmation
Version: canpumf 0.5.3
Affects: get_pumf("Census", "1981 (individuals)")

Summary

For the 1981 Census PUMF (individuals) the metadata shipped by StatCan is self-contradictory. The SPSS command files disagree with the PDF documentation on at least three variables.

canpumf applies a cols_swap override to these three pairs of columns, and warns loudly on every build. The override exists because Statistics Canada's SPSS command file is internally inconsistent: its DATA LIST and its VARIABLE LABELS / VALUE LABELS read the FA* / MA* mnemonics under opposite conventions.

• The record layout PDF reads them as father / mother: FALFACT is the husband's.
• The SPSS labels read them as male / female: MALFACT is the husband's.

The DATA LIST follows the PDF's convention; the label sections follow the other. Same
identifiers, two meanings, one file.

Which spouse each column holds is not in doubt — the PDF, the value-label code
schemes, and the data itself all agree. What is unresolved is what the variables should
be called, and that is a question only Statistics Canada can answer authoritatively.

What canpumf does today

R/registry.R:

"Census/1981/individuals" = .make_entry("Census", "1981/individuals",
  bundle_sps_mask = "ind81",
  file_mask       = "^INDMDF81\\.DAT$",
  data_fixups     = list(
    cols_swap = c(WKACTMA = "WKACTFA", FAOCC81 = "MAOCC81", FALFACT = "MALFACT")
  )),

We swap the column names, then apply the SPSS labels. Functionally this is identical to
following the PDF — reading the PDF's positions and attaching the PDF's prose
descriptions. Every column ends up carrying the right spouse's data either way. The only
difference is the delivered name: we emit MALFACT for position 156, where the PDF
calls it FALFACT.

Each swap emits a warning at build time:

Columns WKACTMA and WKACTFA: names swapped relative to command file —
DATA LIST variable names appear transposed.

The collision, in detail

Statistics Canada's record layout PDF (pumf1981rcl_ev2.pdf, p. 64)

68  1  156      N  FALFACT   Labour Force Activity of H/MLP*
69  1  157      N  MALFACT   Labour Force Activity of W/FLP**
70  2  158-159  N  FAOCC81   Occupation, 1981 of H/MLP*
71  2  160-161  N  MAOCC81   Occupation, 1981 of W/FLP**
72  2  162-163  N  WKACTFA   Work Activity in 1980 for H/MLP*
73  2  164-165  N  WKACTMA   Work Activity in 1980 for W/FLP**

and, spelled out in prose in the variable detail pages (pp. 146, 154):

FALFACT  -  Labour Force Activity of Husband or Male Lone Parent
WKACTFA  -  Work Activity in 1980 for Husband or Male Lone Parent

The PDF is self-consistent throughout: FA* = father = husband/male lone parent,
MA* = mother = wife/female lone parent.

ind81_eng.sps, DATA LIST (lines 69–74)

FALFACT   156 - 156
MALFACT   157 - 157
FAOCC81   158 - 159
MAOCC81   160 - 161
WKACTFA   162 - 163
WKACTMA   164 - 165

Identical to the PDF. So far, no disagreement at all.

ind81_eng.sps, VARIABLE LABELS (lines 172–177)

MALFACT  'HUSBAND/MALE LONE PARENT: LABOUR FORCE ACTIVITY'
FALFACT  'WIFE/FEMALE LONE PARENT: LABOUR FORCE ACTIVITY'
MAOCC81  'HUSBAND/MALE LONE PARENT: OCCUPATION - 1981 BASIS'
FAOCC81  'WIFE/FEMALE LONE PARENT: OCCUPATION - 1981 BASIS'
WKACTMA  'HUSBAND/MALE LONE PARENT: WORK ACTIVITY - 1980'
WKACTFA  'WIFE/FEMALE LONE PARENT: WORK ACTIVITY - 1980'

This is where it breaks. The DATA LIST put FALFACT at 156, which the PDF documents
as the husband's — but this section labels FALFACT as the wife's. The VALUE LABELS
follow the same male/female reading: the label set attached to WKACTMA is the 10-code
scheme, and the PDF documents the 10-code scheme at 162–163, which the DATA LIST names
WKACTFA.

Used as shipped, without intervention, the file therefore attributes each spouse's labour
force activity, occupation and work activity to the other spouse.

The French ind81_fre.sps carries the identical DATA LIST and the same label
assignment, so this is systematic to the release, not a slip in one file.

Which column holds which spouse — not in dispute

Three independent lines of evidence, all agreeing with the PDF:

1. Value-label code schemes

The paired variables use different code schemes. From ind81_eng.sps, WKACTMA (the
husband's, per its label) has 10 codes 0–10; WKACTFA has 12 codes 0–12.
MISSING VALUES declares MAOCC81 (0,17) and FAOCC81 (0,16). Reading
INDMDF81.DAT (486,876 records) at the raw positions:

Positions	Values found	Scheme belongs to	PDF says
162–163	0–10	husband's work activity	husband ✓
164–165	0–12	wife's work activity	wife ✓
158–159	max 17	husband's occupation	husband ✓
160–161	max 16	wife's occupation	wife ✓

The PDF's own code list for WKACTFA (162–163) runs 00–10, confirming the match.

2. Not-applicable agreement (settles the labour-force pair)

MALFACT / FALFACT share one code scheme (0–4), so the code lists cannot
discriminate. A person with no husband has no husband's work activity, no husband's
occupation and no husband's labour-force activity, so the zeros must coincide:

	vs husband's work activity (162–163)	vs wife's work activity (164–165)
column 156	100.00%	92.05%
column 157	92.05%	100.00%

Column 156 is the husband's — as the PDF says. Agreement against the occupation columns
gives the same answer (99.66% / 99.91%).

3. Third-party corroboration (Borealis deposit)

The same PUMF is deposited on Borealis (doi:10.5683/SP3/XHTFC8). Its pumf81i.sps
DATA LIST reads:

    MALFACT    156 -  156        MAOCC81    158 -  159        WKACTMA    162 -  163
    FALFACT    157 -  157        FAOCC81    160 -  161        WKACTFA    164 -  165

— i.e. they made the same choice we did: rename the columns so the SPSS label
sections become correct, rather than rename the labels to match the PDF. Their
pumf81i_codebook.txt documents the variables at those positions with sample frequencies
that match INDMDF81.DAT exactly (MALFACT code 0 = 106,989 at col 156; FALFACT =
81,563 at col 157; MAOCC81 = 105,313; FAOCC81 = 81,109).

This is corroboration of an interpretation, not confirmation of a fact. The Borealis
files are not a Statistics Canada reissue. The codebook header reads:

  Canada. Statistics Canada./
     Census of Canada, 1981 : public use sample tapes user documentation :
     individual file (2 % sample) codebook. Ottawa, Ont. Statistics Canada, September 1982

  Machine-readable DDMS edition codebook compiled by: Canada. Health and Welfare Canada
  Version 1.1, Edited by: Data Library Service, University of Toronto

and pust81i.sas is titled Health & Welfare Canada ed. SPSS control commands. So an
independent data-library team hit the same defect and resolved it the same way — useful,
but it carries no more authority than our own reading.

Notably, the Borealis deposit ships the identical StatCan record layout PDF — the file
is byte-for-byte the same (md5 202e617a4fd0c1172027b422ed8cf303) as the one in the
StatCan bundle. So their modified .sps contradicts the PDF sitting next to it in their
own deposit, unremarked.

What we need from Statistics Canada

1. Confirm the defect. ind81_eng.sps / ind81_fre.sps attach VARIABLE LABELS and
   VALUE LABELS under a male/female reading of FA*/MA* while their DATA LIST uses
   the record layout's father/mother reading. Is this acknowledged?
2. Rule on the naming — the actual open question. For position 156, is the intended
   variable name FALFACT (record layout PDF, FA* = father) or MALFACT (SPSS labels,
   MA* = male)? Both documents are Statistics Canada's own and they disagree. We
   currently emit MALFACT, which means anyone coding from the official record layout
   and asking for FALFACT expecting the husband gets the wife from canpumf. If the
   PDF is authoritative we should invert the override — keep the PDF's names and move the
   labels instead. Same data either way; only the names change.
3. Reissue a self-consistent command file under whichever convention is ruled correct.
4. Confirm scope. We have only examined the individuals file; the households and
   families files ship from the same bundle.

Notes for canpumf

• Only the 1981 individuals file is affected. The 1986 individuals PUMF carries no
  husband/wife spouse-characteristic variables at all, and no other cached census year
  uses cols_swap.
• Until StatCan rules, the naming divergence from the official record layout should be
  called out in the documentation — it silently returns the other spouse to anyone coding
  from the PDF.
• The build-time warning is worth keeping, but its wording ("names swapped relative to
  command file") describes the mechanism rather than the reason, and could point here.

Environment

canpumf 0.5.3
R 4.6.0 (2026-04-24)
macOS 26.6 (darwin 25.6.0)

Start with R/registry.R and inspect the 1981 individuals cols_swap override. Compare ind81_eng.sps and ind81_fre.sps with pumf1981rcl_ev2.pdf, especially the spouse-variable positions and labels. The issue is done when Statistics Canada rules on the naming convention and scope, allowing the override and documentation decision to be made.