Penn Handwerker (1996a, 1998) used factor analysis to test whether the construct of ‘‘domestic cooperation” on Barbados was unidimensional. He asked a random sample of 428 Barbadian women whether their husband or boyfriend helped with any of the following: cooking, washing clothes, washing dishes, bathing children, taking children places, and caring for children. To put these items of domestic cooperation in context, he also asked each woman whether her husband or boyfriend was expected to treat her as an equal and whether her husband or boyfriend did, in fact, treat her as an equal.

Table 22.22 is a schematic of Handwerker’s data matrix. The eight variables are labeled COOK, WASH, DISH, BATHE, TAKE, CARE, EQUAL1, and EQUAL2. Table 22.22 is a profile matrix (see figure 15.2a), but factor analysis is done on a similarity matrix (see figure 15.2b). If Handwerker had asked the women: ‘‘On a scale of 1-5, how much does your husband or boyfriend help you with the cooking,’’ the entries in table 22.22 would have been 1-5. With that kind of data, a factor analysis program would turn the profile matrix into a similarity matrix by calculating Pearson’s r for all possible pairs of columns. But Handwerker asked the women Yes/No questions (that’s why table 21.22 contains only 0s and 1s).

Table 22.22 Schematic of Handwerker's Profile Matrix

ID

COOK

WASH

DISH

BATH

TAKE

CARE

EQUAL1

EQUAL2

1

1

0

1

1

1

0

1

0

2

0

1

0

0

1

0

1

1

3

1

0

0

0

1

1

1

0

438

1

1

0

1

1

0

1

1

SOURCE: W. P Handwerker, adapted from ''Constructing Likert Scales: Testing the Validity and Reliability of Single Measures of Multidimensional Variables,'' Cultural Anthropology Methods Journal, Vol. 8, no. 1, p. 2, 1996a. Reprinted by permission of Sage Publications.

One way to turn a 1/0 profile matrix into a similarity matrix is to calculate the percentage of matches for all possible pairs of columns. That is, when two columns have a 1 or a 0 in the same row, count that as a hit. When two columns have different entries in the same row, count that as a miss. Then, count up all the hits and divide by 428 (the number of possible hits for 428 respondents).

This results in what’s called a simple matching coefficient. (There are other kinds of matching coefficients, but I won’t go into them here.) The result is an 8 X 8 similarity matrix. Table 22.23 shows the results of Handwerker’s factor analysis of that matrix.

In reading the output from factor analysis, we look for items that load high on each factor. For me, that means loadings of at least .60. From table 22.23, it’s clear that the domain of‘‘domestic cooperation” is not unidimensional. In fact, it has three dimensions. Interpreting the results in table 22.23, it seemed to Handwerker that Factor 1 had something to do with ‘‘household chores.’’ He labeled that factor ‘‘Domestic.’’ Factor 2, he thought, comprised ‘‘chores associated with children,’’ so he labeled it ‘‘Children.’’ In open-ended interviews, Barbadian women interpreted the third factor as being about affection and empowerment within families, so Handwerker labeled it ‘‘Affection.’’

Table 22.23 Factor Loadings for Handwerker's Data

Variable

Factor 1 Domestic

Factor 2 Children

Factor 3 Affection

COOK

0.893

0.232

0.161

WASH

0.895

0.121

0.058

DISH

0.824

0.329

0.191

BATHE

0.795

0.324

0.194

TAKE

0.291

0.929

0.159

CARE

0.307

0.922

0.175

EQUAL1

0.188

0.203

0.734

EQUAL2

0.091

0.066

0.854

Explained Variance

39.15

25.516

17.841

SOURCE: W. P Handwerker, ''Constructing Likert Scales: Testing the Validity and Reliability of Single Measures of Multidimensional Variables,'' Cultural Anthropology Methods Journal, Vol. 8, no. 1, p. 2, 1996a. Reprinted with permission of Sage Publications.

Over the next few years, Handwerker had the opportunity to refine and test his scale on two more Caribbean islands, Antigua and St. Lucia. He dropped EQUAL1 because it was redundant with EQUAL2 and he dropped the question about ‘‘caring for children in other ways’’ (CARE) because respondents told him that it was ambiguous.

Handwerker also added four new questions—things that had come up in open-ended interviews as important to women: (1) Does your partner take responsibility for the children for an evening or an afternoon when you have something to do? (2) Does your partner take time off from work to share responsibility for children who are sick? (3) Does your partner talk with you and respect your opinion? (4) Does your partner spend his free time with you?

Table 22.24 shows the results for Antigua and St. Lucia. There are at least four things to notice about table 22.24:

Table 22.24 Handwerker's Data from Antigua and St. Lucia

Variable

Antigua

St. Lucia

Domestic

Children

Affection

Domestic

Children

Affection

COOK

0.793

0.317

0.106

0.783

0.182

0.161

WASH

0.845

0.163

0.090

0.818

0.056

0.018

DISH

0.791

0.320

0.203

0.781

0.197

0.173

BATHE

0.493

0.654

0.207

0.695

0.584

0.090

TAKE

0.271

0.738

0.289

0.246

0.790

0.196

TIME

0.253

0.820

0.258

0.228

0.833

0.061

SICK

0.210

0.786

0.151

0.121

0.742

- .067

FREET

0.100

0.299

0.802

-.010

0.691

0.319

EQUAL

0.147

0.155

0.898

0.177

0.117

0.883

TALK

0.142

0.200

0.883

0.113

0.112

0.909

Explained variance

26.460

26.459

25.051

24.188

27.905

18.191

SOURCE: W. P Handwerker, ''Constructing Likert Scales: Testing the Validity and Reliability of Single Measures of Multidimensional Variables,'' 1996a, CulturalAnthropologyMethods Journal, Vol.8,no.1,p.3,1996a. Reprinted with permission of Sage Publications.

1. Despite the subtraction of some variables and the addition of others, the results from all three islands are very stable. COOK, WASH, and DISH are components of a single large factor across all three Caribbean countries.

2. Although BATHE loads high on the domestic chore factor across all three islands, it also loads high on the children factor for two of the islands. This item should be dropped in the future because it does not reliably distinguish the two factors.

3. FREET loads on the children factor for St. Lucia, but it loads on the affection factor for Antigua. It turns out that Handwerker’s assistant changed the wording for the FREET question slightly when she did the interviews on Antigua. Instead of asking: ‘‘Does your partner spend his free time with you?’’ she asked ‘‘Does your partner spend his free time with you or with your children?’’ (Handwerker 1996a:3). That little change apparently made enough of a difference in the responses to change the factor loading.

4. Across all three replications, the three factors in tables 22.23 and 22.24 account for 70%-80% of the variance in the original data matrix. That is, about three-fourths to four-fifths of the variance in the original data is accounted for by just three underlying variables (the three factors) rather than the full list of original variables. For women across the Caribbean, the construct of domestic cooperation is multidimensional and comprised of three subconstructs: sharing of everyday domestic chores, sharing of responsibilities for children, and affection from men as defined by being treated as an equal.