Straightlining in Web survey panels over time
Straightlining in Web survey panels over time (DOI: https://doi.org/10.18148/srm/2015.v9i2.6128) was written by Matthias Schonlau and Vera Toepoel in 2015. It was published in Survey Research Methods (vol. 9, no. 2).
The authors discuss straightlining and test whether incidence of it increases over time in panel data, as a sort of attrition or panel fatigue.
They use data collected from the (Dutch) LISS panel. This is a probability panel drawn in 2007, with refreshes in 2009 and 2010/2011. Between 6,000 and 10,000 respond to the monthly web surveys. These comprise modules in an annual wave, of which there are 7 in this analysis. They suppress certain models from certain waves, e.g. many modules were not administered to wave 7. The authors also distinguish between modules wherein straightlining is a 'plausible' valid response.
The authors model detected straightlining while controlling for demographic covariates. They use GEE logistic regression so as to:
specify the variance as Φ(p)(1-p)
- "The variance [of a logistic regression] is a function of the mean. Often real life data sets exhibit overdispersion, that is, the observed variance is greater than would be predicted by the model."
- specify a correlation matrix to design clustered variance
The authors do find some evidence of straightlining incidence increasing over time in panel data. There is clearly more incidence in modules with plausible straightlining, so designing surveys with alternating scales is important. They also find that older age is correlated with lower straightlining incidence.