I have been advised to use mixed effects logistic regression (on SPSS) to analyse the results of a psycholinguistic experiment.
Participants are given stimulus sentences containing three nouns or noun phrases (their order varying across different sentence constructions) and asked to write a continuation sentence for each one. Each noun/phrase is defined by the same set of factors. Some factors are categorical, e.g. position in the stimulus sentence (first, middle or last); subject, object or indirect object; proper name or not; qualified by an adjective or not; etc. Some are continuous, e.g. measures of frequency and familiarity.
The objective is to identify the hierarchy of factors that will predict which one of these three 'competitors' will most probably constitute the subject of the continuation sentence. The data for a logistic regression would presumably consist of the factor values of each noun/phrase (IVs) and the outcome (binary DV: subject or not). So three sets of data for each sentence tested.
My concern is whether this approach adequately takes into account the competition between the three nouns/phrases. For example, if only one is a proper noun (name), then will that effect be so strong as to overwhelm all other factors; if two (or three) out of three are proper nouns, then which other factors will be most significant in that context?
In sum, how do I predict the winner? Is logistic regression the most appropriate analysis? Any advice would be very gratefully received. Thanks.
Kevin Glover
PS I reject any suggestion that this is really about gambling! Merry Christmas and goodwill to all.
|