Acquiescence bias

Overview

Some individuals when they complete surveys will tend to agree, rather than disagree, with most of the statements on the survey, such as "Unemployment benefits are too generous" or "The tax rate should be higher for rich people"-called an acquiescence bias. Acquiescence biases can compromise the legitimacy of research. Nevertheless, in some contexts, acquiescence biases can uncover some insightful information about individuals.

Correlates of acquiescence biases

Cognition and intelligence

As Meisenberg and Williams (2008) argued, individuals who demonstrate an acquiescence bias tend to be less educated and intelligent. That is, education and income are inversely related to the inclination to agree with most statements in a questionnaire. Education and income are inversely related to extreme responses as well, which is the tendency of some individuals to strongly disagree or agree with most of the statements on the survey rather than either mildly disagree and agree or feel undecided. Both acquiescence and extreme responses might reflect a neglect of complications.

Personality and beliefs

Individuals who demonstrate an acquiescence bias also tend to be conscientious but prejudiced and intolerant (e.g., Knowles & Nathan, 1997). That is, conscientious individuals tend to comply with instructions, plans, and rules, perhaps preferring to accept rather than question statements. Accordingly, they reject anyone who contradicts their opinions and practices. These individuals do not accept unconventional practices and therefore tend to be prejudiced and racist.

Controlling the effects of acquiescence bias.

When acquiescence biases distort responses, the likelihood of misleading associations between variables increases (Podsakoff, MacKenzie>, Lee, & Podsakoff, 2003). That is, suppose a researcher needs to examine the relationship between two variables, such as attitudes towards capital punishment and the belief that society is fair. A subset of participants might exhibit an acquiescence bias, in which they agree with the majority of questions. As a consequence of this bias, in this subset of individuals, attitudes towards capital punishment might be favorable and society might be perceived as fair. In individuals who do not demonstrate this bias, attitudes towards capital punishment might be unfavorable and society might be perceived as unfair. Accordingly, these variables will seem to be correlated.

Podsakoff, MacKenzie, Lee, and Podsakoff (2003) delineate a set of techniques that can be undertaken to overcome this problem. In this paper, they illustrate these techniques with reference to social desirability biases, but the same approaches can apply to acquiescence bias. For each technique, acquiescence bias must be measured. To measure this bias, researchers can merely determine the average the extent to which participants agree, rather than disagree, with the set of items.

Partial correlation procedures

To control the effect of acquiescence bias, researchers can examine the relationships between variables both before and after controlling acquiescence bias. For example, suppose a bivariate correlation or multiple regression analysis is conducted to examine the relationship between attitudes towards capital punishment and the belief that society is fair. Researchers can then conduct another multiple regression analysis, except the measure of acquiescence bias is included as another predictor or independent variable (Podsakoff, MacKenzie, Lee, & Podsakoff, 2003).

Although effective, this method does present some flaws. First, when this approach is applied, researchers cannot ascertain whether or not acquiescence bias is substantively related to the criterion, such as the belief that society is fair. That is, perhaps individuals who exhibit an acquiescence bias genuinely tend to feel that society is fair. Alternatively, perhaps individuals who exhibit an acquiescence bias do not feel that society is fair, but merely endorse these items regardless because of their tendency to concur with most assertions. In other words, the association might reflect a substantive association or merely a response bias--and this approach cannot distinguish these alternative accounts (see Williams, Gavin, & Williams, 1996).

Accordingly, this approach does not include measurement error in the model (Podsakoff, MacKenzie, Lee, & Podsakoff, 2003). When measurement error is not represented in the model, the coefficients are not estimated as accurately.

Representing acquiescence bias as a latent factors

To represent the measurement error in the model, some researchers apply structural equation modeling and include a latent factor (Podsakoff, MacKenzie, Lee, & Podsakoff, 2003). In this instance, a latent factor, representing acquiescence bias, is included in the model. This approach can be applied only if the researcher is familiar with structural equation modeling. To control acquiescence bias, researchers need to: