Re-visiting Ambivalent Sexism Inventory (ASI): Construct Validity of Benevolent Sexism and Measurement Invariance of ASI
Timuçin Aktan, Bilge YalçındağThe ambivalent sexism theory states that sexism comprises hostile and benevolent beliefs and that benevolent sexism is a second-order factor consisting of protective paternalism, complementary gender differentiation and heterosexual intimacy. The subdimensions of benevolent sexism toward women have recently piqued people’s interest. The Turkish version of the ambivalent sexism inventory’s (ASI’s) construct validity should be reexamined in light of this apparent interest in contemporary studies. Accordingly, in the current study, the aim is to test the preferred structural model in which protective sexism was defined as a second-order factor consisting of protective patriarchy, complementary differentiation between genders and heterosexual intimacy. Moreover, measurement invariance analysis will be used to test the stability of the scale’s structure in different samples. The data of 1803 participants from different studies conducted between 2009 and 2019 (1194 women and 593 men, 16 unidentified) were merged. Findings of the confirmatory factor analyses indicated that the four-factor solution (i.e. hostile sexism and three subfactors of benevolence) fitted the data better than the other models (i.e. one-factor and two-factor models, and the preferred structural model). Explanatory factor analysis via exploratory structural equation modeling revealed a two-factor solution composed of benevolence and hostility, but the findings also underlined two psychometrically weak items. Finally, measurement invariance analyses demonstrated full invariance between private and public university samples, and an invariance between women and men samples except for sample means. Only the means of the samples differed in the women-men comparison, but in a theoretically predicted way, and men had higher scores in all subscales except for complementary gender differentiation. In sum, our findings provided significant support for the construct validity and measurement invariance of ASI while raising questions about the theoretical construct measured and the items needed to be revised.
Çelişik Duygulu Cinsiyetçilik Ölçeğinin (ÇDCÖ) Yeniden Gözden Geçirilmesi: Korumacı Cinsiyetçiliğin Yapı Geçerliliği ve ÇDCÖ'nün Ölçüm Değişmezliği
Timuçin Aktan, Bilge YalçındağÇelişik Duygulu Cinsiyetçilik kuramında, cinsiyetçiliğin düşmanca ve korumacı inançları içerdiği ve korumacı cinsiyetçiliğin korumacı ataerkillik, cinsiyetler arası tamamlayıcı farklılaştırma ve heteroseksüel yakınlıktan oluşan ikinci sıra faktör olduğu belirtmektedir. Kadınlara yönelik korumacı cinsiyetçiliğin alt boyutlarına yönelik ilgi ise yakın dönemde bir artış göstermiştir. Yakın dönem alan yazınında göze çarpan bu ilgi, Çelişik Duygulu Cinsiyetçilik Ölçeği’nin Türkçe versiyonunun yapı geçerliliğini tekrar gözden geçirilmesini önemli kılmaktadır. Bu doğrultuda, mevcut çalışmada , korumacı cinsiyetçiliğin korumacı ataerkillik, cinsiyetler arası tamamlayıcı farklılaştırma ve heteroseksüel yakınlıktan oluşan ikinci sıra faktör olarak tanımlandığı tercih edilen yapısal modeli test etmek amaçlanmıştır. Ayrıca, ölçüm değişmezliği analizi yapılarak ölçeğin yapısının farklı örneklemlerdeki kararlılığını incelemek de amaçlanmıştır. 2009-2019 yılları arasında farklı çalışmalarda yer almış 1803 katılımcının (1194 kadın, 593 erkek, 16 cinsiyetini belirtmeyen) verileri birleştirilmiştir. Doğrulayıcı faktör analizi bulguları, dört faktörlü modelin (düşmanca cinsiyetçilik ve üç korumacılık türü), diğer modellerden (tek faktörlü model, iki faktörlü model ve “tercih edilen yapısal model”) daha iyi uyum gösterdiğine işaret etmiştir. Açıklayıcı yapısal eşitlik modellemesi temelinde yapılmış açıklayıcı faktör analizleri, korumacılık ve düşmanlık şeklinde iki faktörlü bir yapıya işaret etmiş, fakat ölçek içindeki iki maddenin psikometrik açıdan zayıf olduğunu göstermiştir. Son olarak, ölçüm değişmezliği analizleri, devlet ve vakıf üniversitesi örneklemleri arasında tam bir değişmezlik olduğunu; kadın ve erkek örneklerinde ise ortalama farkları dışında değişmezlik gözlendiğini göstermiştir. Kadın ve erkek karşılaştırmasında örneklemler arasında sadece ortalamalar açısından kuramsal olarak beklendik şekilde değişkenlik gözlenmiş ve erkeklerin cinsiyetler arası tamamlayıcı farklılaştırma dışındaki bütün alt ölçeklerde daha yüksek puanlara sahip olduğu bulunmuştur. Genel olarak bakıldığında, bulgular, çelişik duygulu cinsiyetçilik ölçeğinin yapı geçerliliği ve ölçüm değişmezliği ile ilgili önemli bir destek sağlamakla birlikte, ölçülen kuramsal yapının ve ölçek maddelerinin yeniden gözden geçirilmesi ile ilgili konulara dikkat çekmiştir.
The ambivalent sexism theory contributes to our knowledge of sexism by emphasizing a more indirect form of sexism, benevolent sexism (BS), and its relation to hostility toward women, i.e., hostile sexism (HS, Glick & Fiske, 1996; 2001). Early studies put extensive effort on supporting the multidimensional nature of sexism by showing that BS and HS are distinct but related constructs, and that BS has three subdimensions, namely, protective paternalism (PP), complementary gender differentiation (CGD), and heterosexual intimacy (HI, Glick & Fiske, 1996; Glick et al., 2000). Current studies focused on BS and its relations to system justifying beliefs (Bohner et al., 2010; Jost & Kay, 2005). While these studies emphasize the importance of BS, their focus is limited to BS as a single construct, not its subcomponents. More recently, however, few studies indicated that subcomponents of BS have differing relations to sexist outcomes (Oswald et al., 2019; Salomon et al., 2020).
The limited interest in the subcomponents of BS might stem from incongruent findings regarding the construct validity of the ambivalent sexism inventory (ASI). The “preferred structural model” in which BS was defined as a second-order factor was supported in 16 nations, including Turkey (Glick et al., 2000). In the Portuguese and Basque versions of the scale, a four-factor model fitted the data better than alternatives (Costa et al., 2015; Ibabe et al., 2016). Similarly, Sakallı-Uğurlu (2002) observed a four-factor construct for the Turkish version. The present study aimed to reevaluate the construct validity of the Turkish version of the ASI and provide further support for the stability of the observed construct by conducting measurement invariance analyses. Thus, answers to the following research questions were sought:
RQ 1. How well is the theoretical construct validity of ASI?
RQ 2. What is the “natural construct” represented in ASI?
RQ 3. Does ASI have a stabile construct in different samples?
Method
The data of 1803 participants (1194 women, 593 men and 16 unidentified) from different studies conducted between 2009 and 2019 were arranged. Excluding the participants with missing values on ASI final data set resulted in 1633 participants (see Table 1). Participants were asked to report their gender, major, parents’ education level and occupation via a demographic information form. Another scale of the study is the ASI, a 22-item Likert type scale with six response alternatives. Half of the items measure HS. The subcomponents of BS were PP (four items), CGD (three items), and HI (four items). Sakallı-Uğurlu (2002) adapted the scale to Turkish and reported acceptable reliabilities for BC (α =.78) and DC (α =.87).
For the first question of the study, CFAs were conducted to compare alternative models tested by Glick and Fiske (1996) to reveal whether the preferred structural model fit the present data as it did in the original study. For the second question, ESEM was conducted to examine the natural factor structure of the scale. “Psych” package in R was used for these analyses. Finally, “lavaan” and “semTools” packages were used to run invariance analysis for the third question. Following Putnick and Bornstein (2016), measurement invariance was examined by comparing four models: (a) configural, (b) weak/metric, (c) scalar/strong factorial, and (d) strict, residual invariance uniqueness.
Results
CFA findings indicated that a four-factor model had better fit indices compared to alternatives (see Table 2).
The ESEM analyses revealed a two-factor structure (i.e., BS and HS) and raised questions about two items (See Table 3). Item 5 had loadings higher than .30 on both factors, and item 6 loaded on HS, instead of BS.
Findings supported measurement invariance between public and private university samples for 2-Factor and 4-Factor models (see Table 4). Comparing women and men samples, partial measurement invariance was supported for both factorial solutions (see Table 5). In line with the previous studies, men were more sexist in all sub-constructs, except for CGD (see Table 6).
Discussion
The findings of CFAs provided sound support for the construct validity of ASI, and specifically for the four-factor model. Therefore, future studies investigating multivariate relations of the subfactors of BS with sexist outcomes should consider not defining benevolence as a second-order variable in their models (Ibabe et al., 2016).
The natural factor structure of ASI was revealed by ESEM to be made up of HS and BS. However, this research raises concerns regarding the psychometric properties of items 5 and 6. Item 5 may be as strongly associated with PP as it is with HS among younger generations. Item 6 may elicit romantic beliefs rather than sexism. As a result, future research should reconsider the meaning conveyed in ASI items to adapt the scale for younger generations.
Findings related to measurement invariance indicated that 2-Factor and 4-Factor solutions were stable between public and private university samples, and between women and men samples. These findings encourage the fact that BS should be considered in detail. Rather than focusing on as a whole, focusing on the different parts of BS may provide a better understanding of its implications.
