The effects of viewing cute images on the performance of simple computerized tasks in dog owners and non-dog owners

Participants

We used simulation-based power analysis (Lakens & Caldwell, 2021) to find a sample size that will allow us to reach 80% statistical power to find a three-way interaction (Image group (puppies/adult dog images) X Dog ownership (yes/no) X Test (pre/post)) in the alternate task-switching task. For the simulation, we used data from a previous study that we conducted on the same topic (Ziv & Fox, 2021). The simulation showed that the expected effect size for the three-way interaction was ƞ²_p = 0.05 and that 39 participants per group (156 participants in total) were required to achieve 80% power. Therefore, we attempted to recruit 180 participants to account for possible dropouts. In actuality, One-hundred and seventy-eight participants completed the study. The participants were recruited from prolific.co—an online participants’ database. Out of those participants, 164 participants (mean age = 26.06 ± 4.65 years, 114 females) were included in the final analysis (see Data Exclusion section). The participants were randomly assigned to four groups: (1) dog owners viewing images of puppies (n = 35, 24 females), (2) dog owners viewing images of adult dogs (n = 36, 28 females), (3) non-dog owners viewing images of puppies (n = 48, 33 females), and (4) non-dog owners viewing images of adult dogs (n = 45, 29 females). Recruitment to the study was covert based on information the participants provided when they joined prolific.co. To make sure that this information was valid, we also asked, at the end of the study, whether the participants owned a dog. There were some discrepancies between the answers to the study’s questionnaire and the information available on prolific.co. When such discrepancies occurred, we relied on the more recent answers to the study’s questionnaire and therefore, the groups are unequal. Randomization was accomplished automatically by the web-based platform used for the study (Gorilla.sc).

The participants in the preliminary image selection stage were not paid but the participants in the main study were paid £2.5 for their participation. The study was approved by the ethics committee of the Academic College at Wingate (approval # 315), and all participants signed an informed consent form (in the preliminary image selection stage) or completed an electronic informed consent form on the study’s website prior to participation (for the main study).

Tasks

Simon task

The word “right” or the word “left” was displayed on the right or the left side of a central cross. Participants were asked to press the key “j” when they saw “right” (even if it appeared on the left side of the cross) and to press the key “f” when they saw “left” (even if it appeared on the right side of the cross) (Lu & Proctor, 1995; Simon & Wolf, 1963). The words “left” or “right” were shown for 900 msec, followed by 600 msec during which only a central cross was shown.

Alternate task-switching task

A rectangle or square, in a green or blue color, were shown on the upper half or the lower half of the screen. When the shape appeared at the upper half, the participants had to press the key “j” if the shape was green and the key “f” if the shape was blue (irrespective of the shape itself); If the shape appeared at the bottom half, the participants had to press “j” if the shape was a rectangle and “f” if it was a square (irrespective of the color). The shape was displayed for an unlimited time until a key press was recorded. Two shapes were presented at the upper half or at the bottom half of the screen, alternately.

Procedure

The participants performed three blocks of 24 trials in each of the two tasks in a pre-test and in a post-test. Generally, each block took less than 30 or 45 s to complete the Simon task or the alternate task-switching task, respectively. Between pre- and post-tests, participants viewed seven images of either puppies or adult dogs based on group affiliation and were asked to rate them based on their preference, from the image they liked the most to the image they liked the least. We used seven images because previous studies that used the same number of images showed an effect on performance (e.g., Nittono et al., 2012; Yoshikawa, Nittono & Masaki, 2020; Yoshikawa & Masaki, 2021). After the post-test, participants were asked to rate the images on a scale of 0 (not at all) to 5 (very much) for the following criteria: cuteness, infantility, pleasantness, excitement, wanting to get closer. Finally, all participants were asked three questions: (1) how much they loved dogs on a scale of 0 (not at all) to 5 (very much), (2) do they have a dog (yes/no), and (3) do they have another pet (yes/no).

Data exclusion

In accordance with our pre-registration, in the Simon task, we excluded RT values of over 1,000 msec because they were longer than duration of stimulus presentation (900 msec). We also excluded blocks with 13 or less correct responses. These blocks were deleted because they represented less than 55% correct responses suggesting that the participants were not attentive to the task or motivated to perform it according to the instructions (for example, 50% correct responses can be achieved simply by pressing the same key continuously). This led to the removal of three blocks in the pre-test and eight blocks in the post-test. Similarly, in the alternate task-switching task, we excluded blocks with 13 or less correct responses. This led to the removal of 42 blocks from the pre-test and 22 blocks from the post-test. Fourteen participants were removed from the study because they had more than two blocks that had to be removed from one of the two tasks. These exclusions criteria were pre-registered.

Data analyses

Normality was assessed using Kurtosis and Skewness values (values under 2 were considered acceptable). Skewness values were mostly acceptable but in several cases kurtosis values were not. Still, we used parametric statistics because it has been suggested that analysis of variance (ANOVA) is robust to such violations and that type I and type II errors remain constant under different distributions (Schmider et al., 2010). We conducted three-way ANOVAs (Image group (puppies/adult dogs’ images) X Dog ownership group (yes/no) X Test (pre/post)) with repeated measures on the Test factor to assess differences in reaction times and number of correct key presses. To assess the interactions in switch/no-switch trials we conducted two three-way ANOVAs for the pre-test and the post-test: Image group X Dog ownership group X Switch (yes/no). Bonferroni post-hoc analyses and 95% confidence intervals were used for post-hoc testing when necessary. To assess differences in ratings of images’ attributes we used a two-way ANOVA (Image group X Dog ownership Group). In cases where the p value was over 0.05 but under 0.10, and at the same time the effect size was moderate or above (Cohen’s d ≥ 0.5 or ƞ²_p ≥ 0.06), we considered this finding as practically relevant and discussed it as such.

In our exploratory analyses we used Bayesian analyses to support our pre-registered ANOVAs. We also conducted two-way ANOVAs to examine the relationship between the love for dogs and performance. All analyses were performed using JASP version 0.16.2 (JASP Team, 2022) and SPSS 25 (IBM Corp, Armonk, NY, USA). Statistical significance level was set at 0.05.

RT

A three-way ANOVA (Image group X Dog ownership group X Test) with repeated measures on the Test factor revealed a Test effect, F(1, 160) = 13.08, p < 0.001, ƞ²_p = 0.08. The mean RT was 508.79 ± 79.82 msec in the pre-test and 497.83 ± 69.87 msec in the post-test. There was no Image group effect, F(1, 160) = 0.11, p = 0.740, ƞ²_p = 0.00, no Dog ownership group effect, F(1, 160) = 1.31, p = 0.254, ƞ²_p = 0.01, no Test X Image group interaction, F(1, 160) = 2.13, p = 0.146, ƞ²_p = 0.01, no Test X Dog ownership group interaction, F(1, 160) = 0.08, p = 0.771, ƞ²_p = 0.00, no Image group X Dog ownership group interaction, F(1, 160) = 0.04, p = 0.843, ƞ²_p = 0.00, and no three-way interaction, F(1, 160) = 2.13, p = 0.147, ƞ²_p = 0.01.

Correct responses

There were no significant main effects or interactions (all p values > 0.148, all ƞ²_p <= 0.01). The mean correct responses was 22.50 ± 1.07 and 22.54 ± 1.03 in the pre-test and the post-test, respectively (out of 24).

RT

A three-way ANOVA (Image group X Dog ownership group X Test) with repeated measures on the Test factor revealed a Test effect, F(1, 160) = 118.01, p < 0.001, ƞ²_p = 0.42. The mean RT was 1,046.19 ± 280.05 msec in the pre-test and 896.31 ± 205.90 msec in the post-test. There was no Image group effect, F(1, 160) = 0.10, p = 0.751, ƞ²_p = 0.00, no Dog ownership group effect, F(1, 160) = 1.74, p = 0.189, ƞ²_p = 0.01, no Image group X Test interaction, F(1, 160) = 0.65, p = 0.420, ƞ²_p = 0.00, no Dog ownership group X Test interaction, F(1, 160) = 1.67, p = 0.198, ƞ²_p = 0.01, no Image group X Dog ownership group interaction, F(1, 160) = 0.37, p = 0.546, ƞ²_p = 0.00, and no three-way interaction, F(1, 160) = 0.02, p = 0.882, ƞ²_p = 0.00.

Correct responses

There were no significant main effects or interactions (all p values > 0.154, all ƞ²_p <= 0.02). The mean correct responses was 21.41 ± 2.54 and 22.21 ± 1.77 in the pre-test and the post-test, respectively (out of 24).

Pre-test RT

A three-way ANOVA (Image group X Dog ownership group X Switch condition) with repeated measures on the Switch condition factor revealed a Switch condition effect, F(1, 160) = 290.68, p < 0.001, ƞ²_p = 0.64. The mean RT was 1,177.97 ± 302.29 msec in the switch trials and 914.41 ± 290.63 msec in the no-switch trials. There was no Image group effect, F(1, 160) = 0.26, p = 0.608, ƞ²_p = 0.00. There was also no Dog ownership group effect, F(1, 160) = 0.44, p = 0.506, ƞ²_p = 0.01. The mean RT of the dog owners (1,029.84 ± 294.04 msec) was similar to the mean RT of the non-dog owners (1,058.67 ± 269.82 msec) (Fig. 1). In addition, there was no Switch condition X Image group interaction, F(1, 160) = 0.21, p = 0.647, ƞ²_p = 0.00, no Switch condition X Dog ownership group interaction, F(1, 160) = 0.18, p = 0.668, ƞ²_p = 0.00, no Image group X Dog ownership group interaction, F(1, 160) = 0.29, p = 0.589, ƞ²_p = 0.00, and no three-way interaction, F(1, 160) = 0.04, p = 0.839, ƞ²_p = 0.00.

Pre-test correct responses

A three-way ANOVA (Image group X Dog ownership group X Switch condition) with repeated measures on the Switch condition factor revealed a Switch condition effect, F(1, 160) = 4.13, p = 0.044, ƞ²_p = 0.03. The mean correct responses was 10.76 ± 1.27 in the switch trials and 10.64 ± 1.41 in the no-switch trials. There was no Image group effect, F(1, 160) = 1.02, p = 0.314, ƞ²_p = 0.01, no Dog ownership group effect, F(1, 160) = 1.95, p = 0.165, ƞ²_p = 0.01, no Switch condition X Image group interaction, F(1, 160) = 0.06, p = 0.805, ƞ²_p = 0.00, no Switch condition X Dog ownership group interaction, F(1, 160) = 1.55, p = 0.215, ƞ²_p = 0.01, no Image group X Dog ownership group interaction, F(1, 160) = 0.09, p = 0.761, ƞ²_p = 0.00, and no three-way interaction, F(1, 160) = 0.11, p = 0.738, ƞ²_p = 0.00.

Post-test RT

A three-way ANOVA (Image group X Dog ownership group X Switch condition) with repeated measures on the Switch condition factor revealed a Dog ownership group effect, F(1, 160) = 4.15, p = 0.043, ƞ²_p = 0.03. The mean RT of the dog owners (859.08 ± 190.18 msec) was faster than the mean RT of the non-dog owners (924.74 ± 213.79 msec) (Fig. 1). There was also a Switch condition effect, F(1, 160) = 311.68, p < 0.001, ƞ²_p = 0.66. The mean RT was 1,035.12 ± 279.13 msec in the switch trials and 808.66 ± 258.42 msec in the no-switch trials. There was no Image group effect, F(1, 160) = 0.00, p = 0.996, ƞ²_p = 0.00, no Switch condition X Image group interaction, F(1, 160) = 0.13, p = 0.718, ƞ²_p = 0.00, no Switch condition X Dog ownership group interaction, F(1, 160) = 1.48, p = 0.225, ƞ²_p = 0.01, no Image group X Dog ownership group interaction, F(1, 160) = 0.38, p = 0.540, ƞ²_p = 0.00, and no three-way interaction, F(1, 160) = 0.34, p = 0.560, ƞ²_p = 0.00.

Post-test correct responses

There were no significant main effects or interactions (all p values > 0.282, all ƞ²_p <= 0.01). The mean correct responses was 11.14 ± 1.00 and 11.07 ± 0.94 in the switch trials and the no-switch, respectively (out of 12).

Elaborating on differences between dog owners and non-dog owners

In our pre-registered analysis of the post-test, the only significant finding was that dog owners reacted faster than non-dog owners in the alternate task-switching task (in the analysis of the Switch and No-Switch Trials). To assess whether this may have been a spurious finding, we examined this in two alternative methods. First, instead of examining the pre-test and the post-test separately, we conducted a four-way ANOVA (Image group X Dog ownership group X Switch condition X Test) with repeated measures on the two latter factors. There was an expected Switch effect, F(1, 160) = 410.09, p < 0.001, ƞ²_p = 0.72, and a Test effect, F(1, 160) = 118.01, p < 0.001, ƞ²_p = 0.42. However, there were no other significant findings (all p values > 0.119). Specifically, unlike the pre-registered analysis, there was no Dog ownership effect, F(1, 160) = 1.74, p = 0.189, ƞ²_p = 0.01.

Second, we conducted a Bayesian three-way ANOVA for the post-test RT in the alternate task-switching task. This analysis showed that the best model to explain the post-test RT of the alternate task-switching task included the Switch factor and the Dog ownership factor. Compared to this best model (which receives the default value BF₁₀ = 1 in the analysis), the Switch factor alone was second in likelihood (BF₁₀ = 0.64). BF₁₀ values for all other models were under 0.34 (for the null model, BF₁₀ was 4.43E-38). Alternatively, when conducting the Bayesian analysis in comparison to the null model (which receives the default value BF₁₀ = 1), the BF₁₀ for the model that included the Switch factor and the Dog ownership factor was highest (2.26E+37).

Dog loving score and performance

Table 3 presents the distribution of participants across the dog loving scale (0–5). As expected, most dog owners (80%) marked 5 on this scale. Because there were only 18 participants that marked 0, 1, or 2, we collapsed these categories with those who marked 3 and created three groups: those who marked 0–3 (n = 45), those who marked 4 (n = 33), and those who marked 5 (n = 84). We then conducted two-way ANOVAs [Dog loving group X Test] to examine differences in performance variables. There were no significant findings for correct responses in either the Simon task or the alternate task-switching task.

However, for the RTs in the Simon task, the analysis revealed a Dog loving group effect, F(2, 159) = 5.35, p = 0.006, ƞ²_p = 0.06. A Bonferroni post-hoc analysis revealed that participants who marked 0–3 were slower (532.90 ± 87.69 msec) than participants who marked 4 (489.62 ± 54.95 msec) and participants who marked 5 (493.36 ± 66.35 msec). There was also a Test effect, F(1, 159) = 10.67, p = 0.001, ƞ²_p = 0.06, but no interaction, F(2, 159) = 2.97, p = 0.054, ƞ²_p = 0.04.

For the RTs in the alternate task-switching task, the same analysis also revealed a Dog loving group effect, F(2, 159) = 6.53, p = 0.002, ƞ²_p = 0.08. A Bonferroni post-hoc analysis revealed that participants who marked 0–3 were slower (1,073.69 ± 276.30 msec) than participants who marked 5 (927.14 ± 192.67 msec) but not from participants who marked 4 (954.45 ± 209.44 msec). There was also a Test effect, F(1, 159) = 91.60, p < 0.001, ƞ²_p = 0.37, but no interaction, F(2, 159) = 0.29, p = 0.750, ƞ²_p = 0.00.

Bayesian analyses

We conducted three-way Bayesian ANOVAs (Image group X Dog ownership group X Test) to assess the probability of all models compared to the best model. For the Simon RT, the best model only included the Test factor. BF₀₁ was 1.68 for a model that included the Test factor and the Dog owner group factor; 2.54 for a model that included all three main factors; and >4.20 for all other possible models (with most models >10). Similar findings were found for the RTs and correct responses in the alternate task-switching task. For the correct responses in the Simon task, the null model was the best. BF₀₁ was 3.61 for the Dog owner group factor; 3.96 for the Image group factor, and >7.46 for all other factors (with most models >10).

The results of the Bayesian analyses strengthen our pre-registered analyses. Bayes Factors mostly provided moderate to strong evidence that models that include anything other than the Test factor were less likely than the model that includes the Test factor alone. However, there was some likelihood for models that include the Dog owner group as well. This is based on the following interpretation: 3 < BF < 10 can be considered moderate evidence and BF > 10 can be considered strong evidence (van Doorn et al., 2021).