Appendix B: Comparing studies
Mitchell and Edmonson (1999) presented the frequencies of many of the 384 word types used by our participants, and it is these data that Horowitz and Hecht ask readers to compare to a similar combining of their own data. There are problems with such a comparison, for several reasons, some having to do with the extensive differences between studies. We had 46 long interactions; they had 178 very short ones. The individual play interactions in our study lasted 2 to 18 times longer than those observed by Horowitz and Hecht. Our data included the same people playing with a familiar and an unfamiliar dog; theirs included persons playing with a familiar dog only. Our participants were asked directly to play with their own and another dog in whatever way they wanted (but to begin by throwing a ball to make the beginnings of each interaction similar), whereas their participants were self-selected by their desire to share videos of their play with their own dog only.
Another reason that comparisons across studies are problematic is that Horowitz and Hecht, unlike Mitchell and Edmonson, provide only the summed combination of frequencies of words across participants in each study, or in each affect group, to derive their percentages. Such percentages are misleading. A fewpeople might use one word multiple times that no one else used, such that a word can appear with some frequency but not be representative of the group’s speech. For example, the authors include the overall frequency of “kill” as 0.73% of the total wordsin their neutral-play category (p. 782), but this world occurred 7 times in only one episode(of negative-affect play—p. 784).Given that so many words they present occurred with overall frequencies of less than 2%, and given that talk during play can be highly repetitive, it is unclear how common these words were across interactions. Horowitz and Hecht provide no evidence in the form of statistical analyses of their own data to test theirobservations about affect-related words. Perhaps because each person contributed so few words in the 30 to 60 sec videotapes provided (about 31 words per videotape in the positive-affect contexts, and about 14in the neutral-affect contexts—see text), the authors offer only percentages across summed word frequencies, which cannot be examined statistically for differences in word usage between affective contexts. Employing the absolute or (especially) relative frequencies for each word per person are more appropriate measures than those provided by Horowitz and Hecht, and statistical analysis can be performed on such data.
Horowitz and Hecht (p. 780) inaccurately characterize our ideas when they write that “the authors found the words used to be demonstrative of owners’ attempt to ‘control the dog[s]’.” It was not the words used, but the types of utterances (e.g., imperatives, questions, declaratives) as well as other linguistic features (e.g., baby talk) used, that indicated attempts to control the dog or enact other functions. (Surprisingly, Horowitz and Hecht examine none of these features or functions in their data.) The meaning of a word depends on the rest of the utterance it is in; a simple example is “come”:“Come here” vs. “Here I come!”Such is the reason we focused on utterances, rather than words, to determine functions in Mitchell and Edmonson (1999), and why I examined frequencies per person per interaction in the use of types of utterances in later publications (Mitchell 2001, 2004). My interest was to see if the functions of talk to infants, so-called “primary” baby talk, were consistent with the functions of talk to dogs, so-called “secondary” baby talk. Given the importance of utterances and their context to determine word meaning, it is useless to examine words alone in dog-human play to determine their significance. Yet this is exactly what Horowitz and Hecht believe they can do in their assessment of overall word frequency.
Of course some words or word combinations do lend themselves to implied meanings. Participants in the Horowitz and Hecht study seem as a group particularly concerned to show success in their dogs’ play activities, as indicated by the rather high relative frequencies of “good” and “girl/boy” (see their Tables 1 and 2, p. 784)—presumably designating the frequent appearance of “Good girl!” and “Good boy!” “Good” appeared 238 times, and either “girl” or “boy,” 174 times, in the course of their 178 videotapes. Affect matters little in their occurrence: these words were among the 5 most frequent words in both affective contexts. These wordsdid not result from the few (n = 8) instances of tricks the authors observed (p. 782). Theowners playing with their dog in our study also used “good,” but only 63 times across the 24 interactions between familiars spanning an average of 4.5 minutes each. This results in an average of 0.58 “good”s per minute of interaction. If I assume a 45 sec average per videotape (no average time per videotape is provided) for Horowitz and Hecht’s participants, they used more than 3 times that number: 1.78 “good”s per minute of interaction ([238 “good”s ÷ 178 videotapes] ÷ .75 min). The abundance of “good” and “girl/boy” suggests that the citizen scientists who submitted videotapes wanted to show a short snippet of play their dog was good at, rather than what they might do spontaneously or naturally—that is, their data suggest ascertainment bias. Unlike the participants in our study, who attempted and usually failed to verbally control the dog they played with and often found themselves with a dog who wanted to do something different than they did, the dogs in Horowitz and Hecht’s study were apparently already controlled by their owners to do what the owners wanted, and their actions were rewarded with praise for showing that control.