Experiment 1 tested for associations between the side of the QWERTY keyboard on which letters are located and the emotional valence of words that are spelled with these letters, in three languages.
Materials and procedure
We analyzed valence-normed words from three corpora (see Appendix A in the Supplementary online materials): the Affective Norms for English Words corpus (ANEW; Bradley & Lang, 1999), and two translation equivalents of ANEW in Spanish (SPANEW; Redondo, Fraga, Padrón, & Comesaña, 2007) and Dutch (DANEW).
ANEW consists of 1,034 words. Participants used a pencil to rate valence on a 9-point scale composed of five self-assessment manikins (SAMs), which ranged from a smiling figure at the positive end of the scale to a frowning figure at the negative end. Participants were told to mark one of the manikins or a space between two adjacent manikins (see Bradley & Lang, 1999). In SPANEW, translations of the ANEW words were rated by native Spanish speakers using a similar procedure (see Redondo et al., 2007).
DANEW was created for the present study, adapting Bradley and Lang’s (1999) procedure for computerized data collection (see Appendix B in the Supplementary online materials). The 1,034 ANEW words were translated into Dutch by a native speaker. Three of the English words translated to the same Dutch word. Removing these duplicates left 1,031 words in the sample. Each of the participants (N = 132 native Dutch speakers; 14 left-handers, 118 right-handers by self-report) saw 85 of the translated ANEW words intermixed with 74 words from an unrelated experiment, which served as fillers. Participants saw words one at a time and rated them for emotional valence on 9-point SAM scales (ratings for arousal, dominance, imageability, and concreteness were also collected; see Appendix B). Whereas ANEW and SPANEW participants used pencil and paper, DANEW participants responded by clicking one of nine radio buttons located beside the five manikins or between two manikins. In ANEW and SPANEW, the manikins were arranged from left to right. In DANEW, the manikins were arranged vertically on the screen, to avoid any unintended interactions between a left–right rating scale and the left–right positions of the letters that composed stimulus words.
Results and discussion
For each word in the corpus, we computed the difference of the number of left-side letters (q, w, e, r, t, a, s, d, f, g, z, x, c, y, b) and right-side letters (y, u, i, o, p, h, j, k, l, n, m), a measure we call the right-side advantage [RSA = (# right-side letters) − (# left-side letters)]. Overall, there was a significant positive relationship between RSA and valence in ANEW, SPANEW, and DANEW combined, according to a linear regression with items (ANEW words and their translation equivalents) as a repeated random factor using SPSS’s GLM function [b = .044, Wald χ
2(1) = 5.34, p = .02; see Appendix C]. Words with more right-side letters were rated to be more positive, on average, than words with more left-side letters. We call this relationship the QWERTY effect.Footnote 1
To determine whether the QWERTY effect differed across languages, language was added to the regression model as a fixed factor. The mean valence ratings differed between languages, producing a main effect of language [mean valence ratings: Dutch = 5.07, SD = 2.27; English = 5.15, SD = 1.99; Spanish = 4.74, SD = 2.14; Wald χ
2(2) = 101.09, p = .0001]. Importantly, however, language did not interact with RSA to predict valence [Wald χ
2(2) = 0.23, p = .89], and the effect of RSA on valence remained significant when the effect of language and the interaction of language with RSA were controlled [b = .043, Wald χ
2(1) = 5.19, p = .02].
Since there was no significant difference in the strength of the QWERTY effect across languages, an analysis of each separate language is neither required nor licensed. With that caveat, we note that the predicted relationship between RSA and valence was significant in English [b = .043, Wald χ
2(1) = 4.61, p = .03] and in Dutch [b = .051, Wald χ
2(1) = 5.81, p = .02], and a trend in the same positive direction was found in Spanish [b = .035, Wald χ
2(1) = 1.04, p = .31]. It would be inappropriate to interpret these patterns as differing between languages, given the lack of any statistical difference (Wald χ
2 < 1), which cannot be attributed to a lack of power (minimum N = 1,031 items).
A further analysis was conducted to control for possible effects of word length and for the frequency with which individual letters are used in each language (letter frequency).Footnote 2 RSA remained a significant predictor of valence when word length, letter frequency, language, and their interactions were controlled [b = .057, Wald χ
2(1) = 6.95, p = .008].
A final set of analyses tested for effects of handedness in the DANEW raters (no information is available about the handedness of the raters for ANEW and SPANEW). People tend to implicitly associate their dominant hand side of space with positive ideas and their nondominant side with negative ones (Casasanto, 2009, 2011). For this reason, we added handedness to our model to test whether handedness would moderate the effect of RSA on valence. According to a mixed regression model using SPSS’s MIXED function, with subjects and items as repeated random factors, handedness did not interact with RSA to predict valence, F(1, 6077) = 0.16, p = .69, and RSA remained a significant predictor of valence when the effect of handedness and the interaction of handedness and RSA were controlled (b = .061), F(1, 1224) = 4.00, p = .05. Although the QWERTY effect did not differ significantly between right- and left-handers, we conducted an exploratory analysis to determine whether handedness influenced the direction of the correlation between RSA and valence. Right-handers showed a positive association of RSA with valence (b = .060), F(1, 1020) = 4.84, p = .03. Left-handers showed a trend in the same positive direction, which did not approach significance, likely due to the small number of left-handers (b = .022), F(1, 416) = 0.27, p = .61.
Overall, words with more right-side letters were rated to be more positive in meaning than words with more left-side letters, controlling for effects of language, word length, letter frequency, and handedness.