Continuing change in neuroticism during adulthood—structural modelling of a 16-year, 5-wave community study

During the last two decades an increasing interest can be observed in longitudinal personality data and the development of various models that can measure stability and change in these data. In general, test–retest correlations of neuroticism are generally high, but they tend to fall with increasing test–retest intervals, even in adults older than 30. We tested whether this pattern of longitudinal correlations is consistent with the Trait & State (T&S) model, which asserts that the pattern of longitudinal correlations is caused by (i) a common factor (trait); (ii) carry-over effects; and (iii) innovation effects. The T&S model disentangles the variance in neuroticism into an immutable component (trait variance) and a mutable component that changes over time according to a first order auto-regressive process, and that consists of carry-over variance and innovation variance. Two special cases of this model are the T-model (only common factor) and the S-model (only auto-regression). We fitted these models on neuroticism (N) data obtained from a random population sample of 296 subjects in 1970, 1975, 1976, 1984 and 1986. The T-model could convincingly be rejected, but both the T&S- and the S-model fitted well, showing how difficult it is to disentangle trait vs carry-over sources of differential stability. Both models, however, yielded very similar innovation (or change) effects. The standardised 1-year auto-regression was 0.967 (S-model) and the proportion of immutable variance 38% on average (T&S model). The extrapolated 30-year test–retest correlations were 0.37 (S-model) and 0.45 (T&S model). In sum, while our results are not incompatible with an immutable component in individual differences in neuroticism across the adult life span, they did favor the autoregression model. If replicated, our results imply a radical departure from previous research, stressing the need to search for other causes of differential stability than trait factors. The stability found in two-wave studies may have mistakenly been interpreted as reflecting traits.