Search

Sum scores from questionnaire data are frequently analyzed under the normality assumption which is not always tenable due to for instance the skewness of the distribution, or the discreteness of the sum scores for small number of items. Alternatively, an ordinal regression analysis can be applied, but this approach seems appropriate only for very small number of items. The current state-of-the-art approach is to apply a beta-binomial model, though the beta-binomial model does not simply connect to item-response theory models that may describe the underlying items of the questionnaire, and it is more cumbersome to fit to multi domain sum scores or repeated longitudinal single sum scores. This paper proposes a normal-probit-binomial model for repeated sum scores that has an item level interpretation under certain assumptions without having to analyze the individual items. All parameters, the temporal correlation coefficients, and the mean difficulty parameter were almost unbiasedly estimated, irrespective of the set of item difficulty parameters. The coverage probability of the 95% confidence intervals were close to the nominal level, except for the correlation coefficients that were slightly liberal. Our model performed slightly better than the beta-binomial model on cross-sectional data and it handles missing items easily. [ABSTRACT FROM AUTHOR]

Imputation of missing data in high-dimensional datasets with more variables P than samples N, P ≫ N , is hampered by the data dimensionality. For multivariate imputation, the covariance matrix is ill conditioned and cannot be properly estimated. For fully conditional imputation, the regression models for imputation cannot include all the variables. Thus, the high dimension requires special imputation approaches. In this article, we provide an overview and realistic comparisons of imputation approaches for high-dimensional data when applied to a linear mixed modeling (LMM) framework. We examine approaches from three different classes using simulation studies: multiple imputation with penalized regression, multiple imputation with recursive partitioning and predictive mean matching; and multiple imputation with Principal Component Analysis (PCA). We illustrate the methods on a real case study where a multivariate outcome (i.e., an extracted set of correlated biomarkers from human urine samples) was collected and monitored over time and we discuss the proposed methods with more standard imputation techniques that could be applied by ignoring either the multivariate or the longitudinal dimension. Our simulations demonstrate the superiority of the recursive partitioning and predictive mean matching algorithm over the other methods in terms of bias, mean squared error and coverage of the LMM parameter estimates when compared to those obtained from a data analysis without missingness, although it comes at the expense of high computational costs. It is worthwhile reconsidering much faster methodologies like the one relying on PCA. [ABSTRACT FROM AUTHOR]

Sustainable intensification of agriculture requires understanding of the effect of soil characteristics and nutrient supply on crop growth. As farms are increasing in size by acquiring small fields from various farmers, the soil characteristics and nutrient supply might be very different from field to field, while at the same time specific soil properties might limit the nutrient uptake. As a result, there might be a large number of heterogeneous reasons why crop growth varies significantly. New data analysis techniques can help to explain variability in crop growth among fields. This paper introduces Exceptional Growth Mining (EGM) as a first contribution. EGM instantiates the data mining framework Exceptional Model Mining (EMM) such that subgroups of fields can be found that grow exceptionally in terms of three growth parameters (high/low maximum growth, steep/flat linear growth and early/late midpoint of maximum growth). As second contribution, we apply EGM to a case study by analyzing the dataset of a potato farm in the south of the Netherlands. EGM consists of (i) estimating growth curves by applying nonlinear mixed models, (ii) investigating the correlation between the estimated growth parameters, and (iii) applying EMM on these growth curve parameters using a growth curve-specific quality measure. By applying EGM on the data of the potato farm, we obtain the following results: 1) the estimated growth curves represent the variability in potato tuber growth very well (R2 of 0.92), 2) the steepness of the growth curve has a strong correlation with the maximum growth and the midpoint of maximum growth, and the correlation between the midpoint of maximum growth and maximum growth is weak, 3) the subgroup analyses indicate that: high values of K correspond to high maxima; low values of K correspond to low maxima, steep growth curves', and a late midpoint of halfway growth; Mg influences the midpoint of the growth curve; values of B are higher on dry soils with high tuber growth, while low values of B are found on wet soils with high tuber growth; high values of Zn, Mn, and Fe are found in subgroups with low tuber weight, probably related to the soil's low pH. In summary, this paper introduces EGM to obtain understanding in crop response to soil properties and nutrient supply. In addition, EGM provides a way to analyze only small parts of a large dataset, such that the impact of soil factors on growth can be analyzed on a more detailed level than existing methods. [ABSTRACT FROM AUTHOR]

Causal inference from observational data requires untestable identification assumptions. If these assumptions apply, machine learning methods can be used to study complex forms of causal effect heterogeneity. Recently, several machine learning methods were developed to estimate the conditional average treatment effect (ATE). If the features at hand cannot explain all heterogeneity, the individual treatment effects can seriously deviate from the conditional ATE. In this work, we demonstrate how the distributions of the individual treatment effect and the conditional ATE can differ when a causal random forest is applied. We extend the causal random forest to estimate the difference in conditional variance between treated and controls. If the distribution of the individual treatment effect equals that of the conditional ATE, this estimated difference in variance should be small. If they differ, an additional causal assumption is necessary to quantify the heterogeneity not captured by the distribution of the conditional ATE. The conditional variance of the individual treatment effect can be identified when the individual effect is independent of the outcome under no treatment given the measured features. Then, in the cases where the individual treatment effect and conditional ATE distributions differ, the extended causal random forest can appropriately estimate the variance of the individual treatment effect distribution, whereas the causal random forest fails to do so. [ABSTRACT FROM AUTHOR]