In contrast to Second Language Acquisition (SLA) research, the study of individual differences, so far, has not been a prime concern of Learner Corpus Research (LCR). Statistical analyses in LCR have mostly included no more than a single predictor. By assessing the impact of Content and Language Integrated Learning (CLIL) and its selectivity regarding, amongst others, cognitive and affective variables on the passive in corpus data and experimental data, we show that insights gained from such monofactorial approaches are necessarily limited. We formulate regression models containing multiple predictors to overcome these shortcomings, thus showing that data on learner variables can be related to learner corpus data as well as experimental data. Our results suggest that LCR needs to identify further variables which are relevant to explaining variability in text production, but which may not necessarily have received a lot of attention in the experimental approaches adopted by SLA research.