Extended data for 'Ratings of the emotional valence and arousal of collocations and their constituent words'
These materials consist of lists of English collocations along with various associated measures. All these collocations are composed of two words. The most common structures are N-N, Adj-N, and V-‘Any’.
There are two sets of core collocations, one for valence (N = 121) and one for arousal (N = 124). Some lists include calibrator and control items. The two core sets overlap substantially. The ratings of valence and arousal for whole collocations were crowd-sourced using Amazon Mechanical Turk (AMT) (https://www.mturk.com/), The ratings of the constituent words stem from a list compiled by Warriner, Kuperman, and Brysbaert (2013). The list itself is available at:
http://crr.ugent.be/archives/1003
For further details see the main text in the Elsevier journal of applied linguistics, 'System'. The article's title is: 'Ratings of the emotional valence and arousal of collocations and their constiituent words: How can they be useful in L2 vocabulary research?'.
Associated with each core set of collocations is a smaller set of collocations having ratings from both AMT and Warriner et al. These matched, ‘overlapping’ ratings were used to assess the reliability of the new AMT ratings. In the main lists of collocations these ‘overlappers’ are given in red italics. Additionally, there are lists of regression residuals. The closer a residual is to zero, the more accurately rated collocations valence (or arousal) was predicted by the valence (or arousal) ratings of the mean of the constituent word ratings.
Important abbreviations used in the spreadsheet are: AMT = Amazon Mechanical Turk; CW = Constituent word; Geo.mean = Geometrical mean; Harm.mean = Harmonic mean; Most valenced = The CW rating that is the furthest from 5 (i.e., neutral) either toward 1 or toward 9; SD = The standard deviation of the individual AMT ratings obtained for a given collocation; WKB = Warriner et al. (2013); NA = not available.
NA was used in place of values (e.g., CW ratings) that could not be found in the list of WKB. This abbreviation was chosen because the R functions used in the studies can handle datasets that include NA in place of a missing value. For instance, the appropriate calls in base R for calculating Spearman’s and Pearson’s correlations between the variables x and y, when NAs are present, are: cor(x, y, method = "s", use = "pairwise.complete.obs") and cor(x, y, method = "p", use = "pairwise.complete.obs"). The R functions of Wilcox (2012) that were used handle missing values even more automatically. For example, when Wilcox’s R functions are installed (https://dornsife.usc.edu/labs/rwilcox/software/), the call corb(x,y, corfun = spear, nboot = 20000) gives a bootstrap 95% confidence interval for Spearman’s correlation. The call for a median-based linear regression method would be: tsreg(x, y), where x and y are the independent and the dependent variables, respectively.
References
Warriner, A., Kuperman, V., & Brysbaert, M. (2013). Norms of valence, arousal, and dominance for 13,915 English lemmas. Behavior Research Methods, 45, 1191–11207. https://doi.org/10.3758/s13428-012-0314-x
Wilcox, R. (2012). Modern statistics for the social and behavioural sciences. Boca Raton, FL: CRC Press.
