Extreme Response Style

Extreme response style (ERS) is the tendency for survey respondents to answer categorical rating scales in the extreme, end-most intervals, across a wide range of item content. ERS can particularly affect surveys that use Likert and semantic differential scales. ERS is a source of survey error that distorts people's true attitudes and opinions. People with relatively higher ERS will tend to have relatively high or low scores, since they tend to mark extreme intervals, while those with low ERS will tend to have more moderate scores. Thus, apparent differences in survey data and observed scores between people or groups can be an artifact caused by differences in their ERS rather than by differences in their true attitudes and opinions. ERS can also distort the relationship between variables, including survey statistics such as correlations or regression slope coefficients. Distortion from ERS increases when the mean sample response is further from the scale midpoint. ERS is positively correlated with some response styles, such as yea-saying, nay-saying, response range, and standard deviation, and negatively correlated with midpoint responding.

ERS is related to demographic, personality, cultural, and national variables, which makes ERS of particular concern when making comparisons across different countries or cultures. ERS tends to increase with age and decrease with education and household income, or when a person has a more collectivist versus individual orientation. People in Southern European countries tend to have higher ERS than those in Northern European ones. ERS tends to be higher for cultures that are more masculine or that place greater emphasis on differences in power and authority.

ERS depends on characteristics of survey items. ERS tends to be higher when an item is more meaningful to respondents, is worded in the first rather than third person, or written in the respondent's primary rather than secondary language. It can also vary with the scales themselves, such as the number of intervals in the scale.

Several methods have been proposed to measure individuals' ERS and then adjust their observed survey data to compensate, as a means to remove the measurement error induced by ERS. These methods share the common goal of measuring ERS across items probing a range of uncorrelated constructs, to ensure that people's true scores on a particular construct do not unduly affect their ERS scores. One method uses a dedicated battery of items specifically designed and pretested to measure ERS. Other methods allow researchers to use item sets designed for more general survey purposes, provided the items involve several constructs. Several statistical methods have been proposed to isolate observed score variation due to ERS from variation due to differences in attitudes and opinions and other sources of response variance. These methods include structural equation modeling combined with multi-group factor analysis, item response theory, and hierarchical Bayesian ordinal regression.