Using SPSS to Understand Research and Data Analysis.

7.3 Interpreting the Output

The first table of the resulting output file simply lists the number of cases processed (228 total employees). The next table in the output presents the 2 x 2 contingency matrix, or the crosstabulation of frequencies listing the number of male/female employees in the low/high masculinity categories (Figure 7.4). The frequencies highlighted in yellow constitute the four cells of this matrix, and we will need to interpret these if the Chi square test reveals that the variables of gender and masculinity level are significantly related.

For now, looking in the column totals, we see that of the 228 employees 110 are male and 118 female. Looking in the row totals we see that of the 228 employees, 88 are in the low-masculinity category and 140 are in the high-masculinity category. Thus, there is a fairly equal number of male and female employees, and in general, employees are much more likely to be high-masculine than low-masculine.

Examination of the pattern of highlighted yellow frequencies in the cells of the matrix involves:

• comparing the number of men vs. women in within masculinity levels (the horizontal red double-arrows), then
• comparing the number of low- vs. high-masculine employees within gender (the vertical blue double-arrows).

Recall that if these two variables are independent (i.e., not significantly related), then the frequencies would be relatively evenly distributed across masculinity level and gender. That is, there would be relatively equal numbers of men and women in the low-masculinity category as well as in the high masculinity category. Further, there would be an equal number of low-masculine men as high-masculine men, and an equal number of low-masculine vs. high-masculine women.

Scanning these four frequency counts, it appears that the above is generally not true. Rather, there is a pattern of frequencies across the four categories that indicates unequal frequencies. Before we can interpret this apparent pattern, however, we need to examine the Chi square statistic to determine whether or not that this pattern reflects a statistically significant relationship. The Pearson Chi-square value computed by SPSS in shown in the last table of the output (Figure 7.5).

The Chi square value is 13.42. Recall that to determine whether or not this value indicates a significant relationship, we need to examine the probability that this distribution of frequencies occurred by chance alone. Recall that the conventional probability level used to answer this question is .05 and the following decision rule is employed:

• if the probability is greater than .05, then the variables are not significantly related
• if the probability is less than or equal to .05, then the variables are significantly related.

Instead of using the term, probability, SPSS uses the term, significance, and abbreviates it as Sig. More specifically, in Figure 7.5, the Assym. Sig. (2-sided) column lists the probability of interest. The probability is .000 in this column for the Pearson Chi-square statistic.

Note that this means the probability is actually something lower than .0005 (SPSS rounds off to three decimal places). Thus, while the actual probability may not be exactly equal to zero, it is certainly less than the cut-off value of .05. So since the probability is less than .05, we can reject the null hypothesis of chance as an explanation and conclude that there is a statistically significant relationship between gender and masculinity. This justifies our intepretation of the pattern of frequencies in the cells of Figure 7.4. Examination of this matrix in Figure 7.4 reveals that the pattern is more similar to that shown in Table 7.2 than that in Table 7.1.

Starting with the comparisons between men and women (the horizontal red double-arrows), we see that of the 88 low-masculine employees, 59 are women, while only 29 are men. Thus, there were more female EZ employees than male employees in the low masculine category. Further, of the 140 high-masculine employees, 81 are men compared to 59 women. Thus, there were more male employees than female employees in the high-masculine category.

The above pattern follows social stereotypes about sex role identity. That is, given the emphasis on masculinity in the socialization process, it is perhaps not surprising that our sample would have more low-masculine women than men, and more high-masculine men than women.

However, comparisons within genders (the vertical blue double-arrows) reveal a different and interesting pattern. On the one hand, social stereotypes can still be seen for men: of the 110 men, 81 were in the high-masculinity category, compared to only 29 in the low-masculinity category. Thus, the majority of male employees were high-masculine.

On the other hand, the comparison within female employees disconfirms social stereotypes: of the 118 women, half were low-masculine (59) and half were high-masculine (59). Thus, at least at EZ Manufacturing, there were just as many high-masculine as low-masculine women. This illustrates the truism that research and data analyses sometimes confirm hypotheses and expectations, but sometimes they yield surprises - this one reason research and data analysis can be so interesting!

There are a variety of possible explanations of the large number of high-masculine female employees, some of which we will discuss in later chapters. For now, one possible explanation is that EZ is a high-tech manufacturing firm, and technological-orientation is stereotypically associated with masculinity in our culture. Thus, perhaps female EZ employees either adapted their personalities to fit the masculinity stereotype, or perhaps high-masculine women self-selected to work at EZ rather than in more traditionally feminine work roles.

We will gain more insights into the relationships among variables in from our project in Chapter 8, where we introduce a commonly used analytical approach known as correlation. This type of analysis also concerns bivariate distributions, but rather than examining frequencies to assess relationships (as in Crosstabs), correlation directly assesses covariation in scores on two variables.