Response Style Detection and Control

Description

Detecting and reducing systematic tendencies to choose certain response options, such as agreeing, extremes or midpoints, regardless of item content.

Response styles are content-independent habits of answering: acquiescence, disacquiescence, extreme responding, midpoint responding, non-differentiation and others; Roberts lists eight. They distort means, correlations and especially comparisons between cultures or segments. Detection uses either a method factor in confirmatory factor or MTMM models, IRT or latent class models, or representative indicators built from 5 to 14 heterogeneous items per style that share no content (as in RIRMACS). Style indicators can then be entered as covariates or modelled jointly with the substantive constructs. Prevention by design matters as much: item-specific response scales instead of agree-disagree, full labelling, balanced keying, fewer grids and shorter questionnaires. Corrections must be used carefully, because styles vary with content and culture and can remove real differences.

Attributes

Loss type
Siloed
Effort
Days

Steps

  1. Choose the styles of concern for the population and instrument (for example acquiescence, extreme, midpoint).
  2. Design to prevent them: item-specific scales, fully labelled points, balanced positive and negative keying, fewer grids, shorter instruments.
  3. Include 5 to 14 heterogeneous, content-unrelated items per style if using representative indicators.
  4. Detect styles with a method factor (CFA or MTMM), IRT or latent class models, or computed style indices.
  5. Enter style scores as covariates or model them jointly, and compare corrected and uncorrected group differences.
  6. Report which differences survive correction.

Sources

  • The SAGE Handbook of Survey Methodology (Wolf et al., 2016)

Also known as

  • Acquiescence response style
  • Extreme response style
  • Midpoint response style
  • RIRMACS
  • Index of response differentiation

Categories

  • Analytics & Behaviour
  • Survey & Psychometrics