Abstract
We present two openly accessible databases related to the assessment of implicit motives using Picture Story Exercises (PSEs): (a) A database of 183,415 German sentences, nested in 26,389 stories provided by 4,570 participants, which have been coded by experts using Winter’s coding system for the implicit affiliation/intimacy, achievement, and power motives, and (b) a database of 54 classic and new pictures which have been used as PSE stimuli. Updated picture norms are provided which can be used to select appropriate pictures for PSE applications. Based on an analysis of the relations between raw motive scores, word count, and sentence count, we give recommendations on how to control motive scores for story length, and validate the recommendation with a meta-analysis on gender differences in the implicit affiliation motive that replicates existing findings. We discuss to what extent the guiding principles of the story length correction can be generalized to other content coding systems for narrative material. Several potential applications of the databases are discussed, including (un)supervised machine learning of text content, psychometrics, and better reproducibility of PSE research.
Open Scholarship
This article has earned the Center for Open Science badges for Open Data and Open Materials through Open Practices Disclosure. The data and materials are openly accessible at http://dx.doi.org/10.23668/psycharchives.2738 and https://osf.io/pqckn/ and https://github.com/nicebread/PSE-Database.
Notes
1 Affiliation/intimacy is a fusion of originally separate coding systems for affiliation and intimacy. Here we use the abbreviation aff for the combined affiliation/intimacy category.
2 Given that the modeled outcome variable (i.e., raw motive codings) represents strictly non-negative count data, more specific regression approaches would be appropriate. The distributions of raw motive scores of all three motives follow very closely a negative binomial distribution, which suggests a corresponding generalized linear model for count data. However, the main focus of the current analysis is not the hypothesis test, and the residuals at the person level from a Gaussian linear regression correlate ≥ .90 with residuals from a negative binomial regression. Therefore, we focus on the traditionally applied Gaussian linear models and acknowledge the model misspecification, in order to increase the simplicity of practically applying the correction.
3 Again, this applies because we allowed a maximum of one coding per sentence per motive. In the original Winter coding system, multiple codings per motive are possible if two motive images are separated by another motive image within the same sentence.
4 SPSS does not offer a robust regression module, but using the R Essentials plugin, the R function could be used.
5 We had to remove random slopes for pic_position and random effects for study_id to achieve model convergence. Instead we added study_id as categorical fixed effect to control for mean level differences.
6 Descriptive statistics for all pictures, including the new pictures, are at https://osf.io/pqckn.