Handling missing data

Missing data is a problem in all quantitative investigations. In any survey, for example, some respondents fail to answer some items. The best defense against this problem is to take all possible steps to get valid responses to all items from all respondents, but some omissions are inevitable. What do you do in these situations?

There is no satisfactory answer to this problem, but there are several ways of dealing with it. One solution is simply to eliminate any cases for which data are missing for some indicator. If you have to drop too many cases, however, this should be a warning that your data are an insufficient basis for constructing an index. Another solution is to make an "educated" estimate about what the missing value should be. In our example, knowledge of the community might suggest a basis for supplying a value for the missing data. Still another solution is to create a substitute for the missing data. You can find the average of the values for the indicator for which there is missing data and substitute this value for any missing response. Using the average will have the least distorting effect on the remaining data.

The problem of missing data becomes more serious when you are using only a few indicators and have 100 or fewer cases. When you have less data, any loss affects the final results much more than when you have a number of indicators and a large number of cases.

Validation of composite measures

The item analysis we described earlier also provides a form of internal validity. When the items show moderate association with each other, we have some evidence that they are valid measures of the dimension or variable being measured. External validity tests provide an additional and stronger basis for demonstrating the validity of an index. One of the techniques described in Chapter 6 should be used to estimate the validity of a composite measure.

Typologies

A typology is a form of composite measurement, but typologies differ from scales and indexes in two important ways. First, while scales or indexes are unidimensional, meaning they measure only one dimension of a variable, typologies are frequently multidimensional: They are based on a combination of different characteristics or variables. Researchers use typologies when they want to describe persons, groups, or other units of analysis in more than one dimension at the same time.

Swanjord (1988) created four categories for describing how Kuwaiti women were portrayed in publications over a sixty year period, from the early 1920s through the early 1980s. Each category was defined in terms of a number of attributes writers used in describing Kuwaiti women. The four categories or types were: traditional, fundamentalist, secular, and modern. Each represented an abstract type of woman as described in various publications. Together, the four types formed a typology for analyzing the contents of the publications. Swanjord found that portrayals of the women were predominately fundamentalist or traditional from 1920 through 1980 and shifted to secular portrayals in the 1980s. The shift reflected the social changes in Kuwait following the large increase in revenues from the sale of oil.

Noting that the value of women's work in developing countries is often overlooked and under valued, Donahoe (1999) set out to more fully measure the value of their work. In the course of her research with a sample of Egyptian households, she developed a five-point typology of women's work. She classified women as: (1) employees, employed in some work not in the family or with relatives; (2) income generators, employed in some familial arrangement; (3) providers, working at least 14 hours per week in subsistence production; (4) part time providers, same as providers, but working less than 14 hours per week; and (5) housewives, doing none of the above. Donahoe argued that her typology accounted for a larger amount of the productive work done by women than is generally recognized.   

Typologies are useful for measuring independent variables. In Swanjord's investigation, the typology for images of Kuwaiti women was used to analyze the trend in images used to describe women and other variables. Donahoe's typology can be used as a nominal variable for the study of women's work in any developing country. Because they are usually multidimensional, typologies generally are not be used as dependent variables.