Fig 1.
Illustration of multivariate normative comparisons in a situation with scores on two neuropsychological tests.
The double-headed arrows denote the 95% univariate intervals. The ellipses denote the 95% multivariate region. The dots denote the mean score in the norm group. The triangles depict a patient’s scores. In the left panel, tests are uncorrelated (r = 0.0). In the right panel, tests are correlated (r = 0.7).
Table 1.
Example of a missing data pattern, where 1 = available, and 0 = missing.
For each test, and each study, there are scores missing, although all test co-occur at least once.
Table 2.
Simulated example of a multilevel dataset with one row per test score.
study indicates study number; ID indicates participant number; age, gender and education are background variables; z(1), z(2) and z(3) are indicator variables; test indicates test number and score indicates the score on the test with that number.
Table 3.
Model specification for a multilevel model with three tests, and three background variables (age, gender and level of education), including specification of between and within study covariance structures.
Table 4.
Settings used in the two simulation studies.
Fig 2.
Missing data patterns for the 0%, 40% and 70% missing data conditions, with studies on the y-axis and tests on the x-axis.
Colored boxes are non-missing test scores, white boxes are missing test scores.
Table 5.
Parameter values in the two simulation studies.
Fig 3.
False positives (where number of deviations = 0) and sensitivity (where number of deviations > 0) as a function of the number of simulated deviations, for 0%, 40% and 70% missing data in the norm group.
Error bars represent 95% confidence intervals.
Fig 4.
Four selected bivariate plots.
The ellipses denote the 95% multivariate region. The dots denote the mean score in the norm group. The triangles depict the patient’s scores.