Sentence examples for substantive traits from inspiring English sources
The advantage of modeling all attitudinal scales simultaneously when making this assumption is that it provides more reliable estimation of θ ERS, and thus better control of ERS with respect to measurement of the substantive traits (see Bolt & Newton, 2011).
Such curves play a useful role in understanding why controlling for ERS introduces larger changes in the mean estimates of the substantive traits for some countries but not others.
The second level of the model, the country level, associates with each country a mean vector, μ g and covariance matrix ( {varSigma}_g^{(1)} ), representing the mean and covariance matrix across the substantive traits and ERS.
Bolt & Newton (2011) also showed how this general approach can be extended to take into account multiple scales that use the same response format but measure different substantive traits.
As noted, the model emphasizes the simultaneous influence of both the substantive trait and response style on a respondent's selection of extreme versus less extreme categories.
Intuitively, one might expect the bias to be negative when substantive trait levels are low (as ERS leads to selection of more extreme lower scores than otherwise expected) but positive when substantive trait levels are high (as ERS lead to selection of more extreme higher scores).
As such, it accounts for the possibility that respondents and/or countries that are selecting a large portion of extreme responses may be doing so in large part because they have extreme levels on the substantive trait.
Another alternative is the simultaneous administration of other self-report scales involving heterogeneous content, making ad hoc indices of response style less susceptible to influence of the substantive trait.
Relative to earlier studies on response styles which generally only verified the existence of response styles, the MIRT model is also capable of correcting ERS bias in providing substantive trait estimates.
A less than perfect correlation is expected to the extent that the current model assumes extreme responses are in some cases caused by the level of the substantive trait.
Importantly, such a correction assumes that the biasing effects of response style are constant across levels of the substantive trait, an assumption that, as seen below, is sharply at odds with the current model based approach.
