Weinrich, James D., Snyder, Peter J., Pillard, Richard C., Grant, Igor, Jacobson, Denise L., Robinson, S. RenŽe, McCutchan, J. Allen (1993) A factor analysis of the Klein sexual orientation grid in two disparate samples. Archives of Sexual Behavior 22: 157-168.

INTRODUCTION

In the eternal drama of the lumpers versus the splitters, the concepts of sexual orientation and sexual preference have recently occupied center stage, and the Klein Sexual Orientation Grid has been both hailed and damned by the critics. This instrument was first sketched by Klein (1978, pp. 17-19) in a popular book. It was developed more fully in a magazine article 2 years later (Klein, 1980), and finalized in a scholarly journal some time after that (Klein et al., 1985). It stands as middle ground between those who feel comfortable using a single word (homosexual vs. heterosexual) to describe their subjects and those who feel that a complete clinical sexual history must be taken in order to understand the most important aspects of a person's sexual life. It is the purpose of this paper to apply actual data to this question, and thereby help investigators decide the appropriate level of detail with which to describe their subjects and patients.

We begin with a review of the Klein Grid's predecessors. The Kinsey group made the first advance over the earliest popular categorization (in binary fashion as homosexual or heterosexual) by proposing the heterosexuality/homosexuality rating scale (Kinsey et al., 1948, p. 638). This scale is now commonly called the "Kinsey scale." Kinsey and his co- workers have admitted that they actually considered two parameters in their ratings--"overt sexual experience" and "psychosexual reactions"--but claimed that "in the majority of instances the two aspects of the history parallel." When they do not, "the rating of an individual must be based upon an evaluation of the relative importance of the overt and the psychic in his history" (all quotes from Kinsey et al., 1948, p. 647). Indeed, later workers at the Kinsey Institute specifically rated their subjects on these two scales (e.g., Bell and Weinberg, 1978, p. 54), but they did not present their data separately by the two parameters, since the two values were only slightly different in most cases. Bell and Weinberg used the sum of the two separate Kinsey ratings in their analyses, and called an individual "homosexual" if he or she had a summed score of 4 or more. They reported that discrepancy between the two ratings "was quite rare among the 'homosexual' group ... nearly 90 percent of our 'homosexual' respondents (except for the black lesbians' 77%) scored 10 or more--i.e., predominantly or exclusively homosexual in both feelings and behaviors" (Bell and Weinberg, 1978, p. 35).

With the advent of the bisexual movement and widening appreciation of dimensionality issues in gender research--such as the two-dimensional nature of the Bem scale (Bem 1974, 1981)--several investigators have discussed whether sexual orientation and preference ought to be measured on two or more independent scales. The scheme that has received the most attention is Klein's grid (cited above), at three different time periods: Present, Past, and Ideal (or Future), for a total of 21 measurable dimensions.

With the understanding that no measure of sexual orientation or preference can please every investigator, the factor structure (or principal components) of the Klein Grid remains unknown; apparently, no such factor analyses have been reported in the literature. Luckily, two of the Klein Grid's seven areas approximate the two Kinsey scales that had been in use informally, so a factor analysis of the Klein Grid would contribute to an understanding of the Kinsey scale too. In particular, it would help validate or challenge the earlier data that suggested that not much is lost in most samples by considering fantasy and behavior to be two aspects of a single phenomenon.

There are other questions in sexology other than the purely technical ones which can be addressed by factor analysis. For example, some have been interested in developing an analysis of human sexual arousal by categorizing the patterns of sexiness people recognize in themselves as "limerent" or "lusty" (Weinrich, 1987, chap. 6; 1988). This dichotomy corresponds roughly to "love" vs. "lust," and was arrived at specifically because a bisexual man challenged the senior author to develop a theory in which loving sexiness was completely integrated with other kinds of sexiness. Two or three of the Klein Grid's seven domains could be interpreted as addressing a similar distinction.

And finally, since factors are not really a property of a set of questions but rather of how those questions were answered by a particular population at a particular time, it is crucial to look at the factor structure in at least two different samples. Such an opportunity presents itself here.

OBJECTIVES

The senior author has participated in two moderate-size studies in which the Klein Grid was administered to subjects. One study's sample included a higher proportion of self-identified bisexuals than is commonly obtained in sex research (1/3 of a total of 90). The other study, of the human immunodeficiency virus (HIV-1), uses a sample that is heavily homosexual and somewhat bisexual in composition, with a smattering of heterosexual respondents. These samples are described in detail below.

The objectives of the present study include (i) The discovery of the principal components in the 21 items that constitute the Klein Grid. (ii) The analysis of these components to see if any of them correspond to the terms "sexual orientation," "sexual preference," "love," or "lust." (iii) The analysis of these components in a least two disparate samples to see if the components identified are similar or dissimilar by sample.

METHODS

As mentioned above, two samples were available to the senior author for analysis. The first was recruited in a large Eastern city and the second on the West coast.

The Fat Metabolism Sample

This sample consisted of 90 men who responded to advertisements in local weekly newspapers (gay and nongay), asking for volunteers for a study of fat metabolism. Advertisements were worded identically, offered a $10 payment, and did not mention sex research or that there were any sexual orientation requirements for the volunteers. Potential participants were invited to the research laboratories, at which time they were informed of the nature of the study (including its sexual aspects), signed a statement giving informed consent, and completed a brief questionnaire with an item asking for one's self-identification as homosexual, bisexual, or heterosexual.

Responses to this question were tabulated frequently as recruitment progressed, with the aim of recruiting a sample evenly divided among the three broad categories of sexual-orientation self-identification. The heterosexual and homosexual categories were filled at about the same time, by which time only 7 bisexual slots remained unfilled. At that point, all ads were changed to mention that only bisexual respondents were needed, and the remaining positions were quickly filled. The final sample consisted of 31 heterosexual, 30 bisexual, and 29 homosexual men, as defined by their own answers to the single self-classification question, and without regard to clinician impressions or the subjects' responses to other Klein Grid or Kinsey scale items on the questionnaire.

The HIV Sample

This sample consisted of 78 homosexual and bisexual men who were enrolled in a large AIDS research center. Although the number of subjects will ultimately become much larger, these 78 had been recruited and administered the Klein Grid at the time of data entry. Roughly 25% of the men were seronegative controls; the rest were seropositive at various stages of HIV disease.

Administration

Both samples were administered written versions of the Klein Grid, taken directly from the scholarly presentation of that scale (Klein et al., 1985). Data were entered into computers and then analyzed using the StatView program. The factor procedure was principal component analysis, and the transformation method was orthotran/varimax. The factor extraction method was the default for the StatView program, which involves the combination of two criteria--namely, the number of factors extracted is the larger of the numbers determined by a 75% variance rule (stop adding factors once the proportionate contribution of the eigenvalues exceeds 0.75) and a root curve analysis (stop at the eigenvalue associated with the point of inflection of a plot of the eigenvalues from largest to smallest).

RESULTS

Means, Standard Deviations, Intercorrelations

To facilitate comparisons between and within the two samples, we have provided the basic descriptive information on each of the 21 Klein items separately for the two samples in Table I, which also gives simple intercorrelation of each of the 21 items with the others. Readers may wish to consult this table to informally validate the results of the factor analyses reported below.

[TABULAR DATA OMITTED]

Fat Metabolism Sample

Each of the 21 Klein variables had variable sampling adequacies of 0.8 or more, which comfortably exceeds the usual minimum requirement of 0.5, and the total matrix sampling adequacy was 0.92. This suggests that the Klein items represent, as desired, a homogeneous collection of variables suitable for factor analysis. Bartlett's test of sphericity, chi sq.(230) = 3514.14, p < 0.0001, strongly implied that the correlations among the various variables are significantly different from zero. Six eigenvalues were derived by the program, of which the first three were greater than 1. The factors associated with these eigenvalues were associated with 69, 9, and 5% of the variance.

The factor loadings for the orthogonal solution (after rotation) are presented in Table II (first three columns). All of the 21 Klein items loaded on Factor 1 at 0.55 or higher. It was, accordingly, named "General Sexual Orientation." Factor 2 was associated with the three items relating to the sex one prefers to socialize with and the three items relating to emotional preference (at between 0.4 and 0.6 in all cases). It was named "Socialize With/Emotional Preference." The three items concerning one's emotional preference were singled out by Factor 3, which was named "Emotional Preference."

When the orthogonality requirement was relaxed, an oblique solution primary pattern matrix was obtained. An oblique solution tends to more clearly separate the loadings of items on overlapping factors. These loadings are presented in Table III (first three columns). All of the Klein items except emotional preferences and socializing preferences loaded on Oblique Factor 1. The three "socializing with" variables loaded on Oblique Factor 2, while the three emotional preference items loaded on Oblique Factor 3.

As would be expected with results such as these, the three oblique factors were intercorrelated. As shown in Table IV (first three columns), all three pairs of intercorrelations were approximately 0.55.

HIV Sample

Each of the 21 Klein variables had variable sampling adequacies of 0.68 or more; the total matrix sampling adequacy was 0.86. Bartlett's test of sphericity, chi sq.(230) = 2847.31, p < 0.0001, implied that the correlations among the various variables are significantly different from zero.

As for the other sample, six eigenvalues were derived by the program, of which the first three were greater than 1. The factors associated with these eigenvalues were associated with 60, 12, and 8% of the variance.

The factor loadings for the orthogonal solution (after rotation) are presented in Table II (last three columns). All of the 21 Klein items loaded on Factor 1 (General Sexual Orientation) at 0.44 or higher; 18 of the 21 items loaded at 0.69 or higher. Factor 2 (Emotional Preference) related to the three items relating to the sex one prefers to socialize with (these items loaded at better than 0.65 in each case). Factor 3 was difficult to interpret at this stage, although all seven "Ideal (or Future)" items loaded negatively, unlike the other 14 items, and usually substantially so. Naming of this factor was reserved for the oblique analysis.

[TABULAR DATA OMITTED]

The factor loadings for the oblique solution (after rotation) are presented in Table III (last three columns). The striking event at this stage was the emergence of Oblique Factor 3 as a clear "Ideal Versus Past/Present" factor; six of the seven "Ideal" items (all except Socialize With) loaded heavily here. It might also be termed the "Satisfaction/Dissatisfaction with Past and Present" factor. Factor 2 remained a Socialize With factor (with notable but lesser loadings by the three Emotional Preference items and the three Life-style items). Factor 1 remained the General Sexual Orientation factor (with the major items from Factors 2 and 3 removed).

Again as would be expected with results such as these, the three oblique factors were intercorrelated. As shown in Table IV (last three columns), the three pairs of intercorrelations ranged between approximately 0.4 and 0.7. The correlation between Oblique Factors 1 and 3 seemed higher than the other two.

DISCUSSION

The similarities and contrasts between the results in the two analyses are striking. Most clearly, both analyses' most important and first-emerging factor was one which was associated with the majority of the 21 Klein items--all 21, in fact, in the orthogonal analyses. This fact suggests that for these samples (at the very least), there is such a thing as sexual orientation and it is legitimate for people to use the term in a way that encompasses a wide variety of aspects of life. In this respect, the lumpers win.

In both samples, however, there was support for other hypotheses. Who one socializes with turned out to be an important additional piece of information in both samples. This lends support to the intuition of many (the splitters among them) that this set of items is something rather different from sexual orientation.

There are many questions that may profitably be addressed by the use of more advanced statistical techniques (which will be pursued after the accumulation of more data). For example, is three the correct number of dimensions with which to explain these data? In both samples? LISREL or EQS could be used to address these questions as well as many others of interest.

The fact that the Klein Grid factored in similar ways in both samples (for the first two factors) suggests that the grid appears to be an externally valid measure. Remember that the first sample was evenly distributed across the categories of heterosexual, bisexual, and homosexual, whereas the second was overwhelmingly gay-identified, so this result is more important than it might at first appear.

The differences in the two triplets of factors are equally intriguing. Mild support for the notion that love as opposed to lust is an important dimension independent of sexual orientation was found in the Fat Metabolism sample, and to a lesser extent in the HIV sample. This may have been expected, since bisexual men were more common in the first sample, and are more likely to mention the importance of love as opposed to lust in their relationships with women. Anecdotally, the senior author has observed that one of the most common combinations of experiences causing men to identify themselves as bisexual is when a man with erotic arousal patterns to other men discovers himself in love with a woman. Perhaps this kind of bisexual man is predominantly responsible for the salience of these second factors.

Likewise, the fact that so many of the Ideal ratings loaded on the third oblique factor emerging in the HIV sample suggests that conflict over one's ideal versus actual feelings and behaviors may be important to study in certain populations. Gay men who carry HIV may be more likely to regret their homosexuality and therefore have an Ideal orientation that is significantly different from what their Present and Past patterns have been. In the HIV sample, this hypothesis is not borne out on the sample as a whole. However, when factor score weights were applied to each subject's answers to generate scores for each subject on the three factors, those who scored highest on that sample's Factor 3 (Ideal vs. Present and Past) are nearly all seropositive (data and analyses not presented here).

The factor patterns obtained in this study suggest that a clustering procedure could be used to discern sexological typologies. For example, as indicated in the previous paragraph one might be able to correlate scores on HIV-sample Oblique Factor 3 with other variables of interest, perhaps a DSM-III diagnosis of ego-dystonic homosexuality or personality traits not so clearly upsetting or pathological.

Another clustering could be grouped around a two-way classification on the basis of Factors 1 and 2; this should identify men who are congruent or discordant on the basis of General Sexual Orientation and on Social/Emotional Preference, a classification that is expected to be important if Weinrich's (1988) limerent/lusty theory is correct. Preliminary analysis suggests that discrepancies are common in both samples when the General Sexual Orientation factor is tabulated versus Factor 2. In the Fat Metabolism sample the majority are discordant, whereas in the HIV sample the majority are concordant. These leads will be pursued in further research.

Finally, what did not emerge from the analysis may be just as important as what did. In particular, no factors emerged that separated Present from Past items. This suggests that in these samples, changes in sexual orientation are rare, or occurred before the 1-year horizon that respondents are asked to use in separating Present from Past preferences and behaviors. This result is in striking contrast to analyses performed by Klein et al. (1985, p. 43), who found that there was "a significant difference between the Present scale and the Past scale, but none between the Ideal and Present scales" in their sample as a whole--which spanned, however, all three major sexual-orientation groups. When this was examined separately in their homosexual, bisexual, and heterosexual groups, a considerably more diverse pattern emerged.

In summary, both the lumpers and the splitters can find comfort in the results from our factor analyses. Lumpers will delight in the finding that all 21 Klein items loaded onto the first orthogonal factor in both samples; they will see in this finding the suggestion that little violence is done if one reduces the Klein Grid to a single number called Sexual Orientation. Splitters interested in the subtler second- or third-degree approximations will delight in the finding that social and emotional preferences are discernibly different enough to emerge in analyses of two different samples, and will urge more studies on larger samples to see if a larger N will bring out even more detail. And we, as scientists on the fence, are happy to see the debate continue.

REFERENCES

COPYRIGHT © 1993 Plenum Publishing Corporation