Data from two participants could not be analysed because they did not return for the second session. The final sample size was 29 participants (16 females).
Water consumption and hydration status effects on thirst and mood scales
In the water condition, participants drank a mean of 303.44 ml (SD 158.21; range 50–500 ml). To test whether people were indeed dehydrated in the no water condition after test as well as on both mornings, we ran a 2 (water vs no water) by 2 (waking vs end of test) ANOVA on osmolality readings. There was no effect of day, F(1,28) < 1, but a main effect of test time, (F(1,28) = 5.96, p = .021, ηp2 = 0.176), as well as an interaction, (F(1,28) = 6.231, p = 0.019, ηp2 = 0.182). Whereas there was no difference between hydration readings in the water condition before (M = 735, SD = 252) and after (M = 758, SD = 235) testing, there was a difference in the no water day, with readings lower before (M = 693, SD = 218) than after (M = 813, SD = 217) testing (see Fig. 1). This suggests that on a group level, participants were reasonably dehydrated (osmolality readings of ca 700–800 mOsmo/kg), but also that in the no water day, the dehydration became significantly worse during the morning compared to the water day (Edmonds et al., 2013). Thus, water supplementation on the water day prevented further dehydration, which seemed to happen on the no water day as testing went on through the morning. Thirst ratings also confirmed that participants arrived thirsty: participants rated themselves as having greater subjective thirst on the occasion that they were not offered water (F(1,27) = 46.112, p < 0.001).
The responses to the I-PANAS-SF mood scale were mostly unaffected by water supplementation, thirst, order and osmolality. There were two exceptions to this statement: there was a water supplementation x order interaction for “attentive” and an osmolality effect on “inspired”. Participants who received water in their first session reported being more “attentive” on that occasion compared to their second session in which they did not have any water (F(1,27) = 16.00, p < 0.001). In the case of urine osmolality, dehydrated participants (as evidenced by higher urine osmolality) rated themselves as significantly less “inspired” (F(1,27) = 4.276, p = 0.048). There was no effect of thirst (high vs low scorers) on any of the items presented in the I-PANAS-SF mood scale.
Water consumption effects on executive functions
Mean scores on CANTAB tests were screened for normal distribution and outliers, using the interquartile range rule of g = 3 (Hoaglin et al, 1986). Only one RVP errors data point was substituted with RVP misses in one condition for one participant who had a very high RVP false alarm rate in one condition. For all other participants, the RVP total errors were calculated as the sum of the number of false alarms and number of misses.
Performance on each of the CANTAB tasks was analysed using mixed-design ANOVAs, one separately for effects of order, thirst, and osmolality. The within-participants factor in each ANOVA was water supplementation (water vs no water). The between factors in the respective ANOVAs were order (water first session or water second session), hydration [osmolality: high > 827.5 mOsm/kg or low < 827.5 mOsm/kg; i.e., dehydrated (15) or hydrated (14)], and finally thirst (high or low after median split; 14 participants classified as thirsty and 15 as not thirsty). There were no significant effects or trends for the factor order or water (see Tables 1, 2, 3), bar two exceptions. There were trends for ChoiceRTs to be generally faster in the water conditions (p values between 0.066 and 0.073; Tables 1, 2, 3), and there was a significant interaction for water and order in the RVP tasks (Table 1), with more errors in the water condition (M = 17.80, SD = 4.72) compared to the no water condition (M = 21.13, SD = 5.55) when participants received water in their first session, p = 0.007, but vice versa when they received it second, p = 0.057 (water: M = 22.07, SD = 3.15; no water: M = 20.50, SD = 3.65). Therefore, hypothesis 1 could not be retained.
Table 1 CANTAB test means, SDs and F ratios by water condition (water/no water) and order (water first/no water first)
Table 2 CANTAB test means, SDs and F ratios by water condition (water/no water) and post-testing urine osmolality (low/high) as measured on the day participants did not receive any water
Table 3 CANTAB test means, SDs and F ratios by water condition (water/no water) and thirst (low/high) as measured on the day participants did not receive any water
Regarding the main effects of between-subjects variables, there was a marginal effect of order on the ChoiceRT, F(1,27) = 4.034, p = .055, with higher RTs in the first session (M = 327, SD = 12) than in the second (M = 290, SD = 13). Otherwise, there were no effects, all Fs < 1, except for IED errors and order, F(1,27) = 1.503, p = .23, and order effects on RVP error rates, F(1,27) = 2.161, p = .153. Controlling for the amount of water each participant drank (using ANCOVAs) did not change the pattern of effects, all Fs(1,28) < 1, except for a similar trend as above for ChoiceRT F(1,28) = 3.53, p = .71. Additional ANCOVAs using thirst and osmolality as co-variates (rather than median split as a between-group factor), again found similar effects for the independent variables on CANTAB scores, all Fs < 1 (except the trend of water for ChoiceRT, with p values between 0.065 and 0.071.
Water consumption effects on judgment and decision-making performance
Three mixed-design ANOVAs were performed with total correct score for the combined judgement and decision-making tasks (six heuristic vignettes and three CRT vignettes in each session, see Methods) as the dependent variable, analogous to the ANOVAs for the executive function tests above. The within-participant factor in each ANOVA was water supplementation (water vs no water). The between factors in the respective ANOVAs were order (water first session or water second session), hydration (osmolality: high or low; i.e., dehydrated or hydrated), and finally thirst (high or low after median split). In all three ANOVAs, there was a main effect of water supplementation (Table 4), for the ANOVA on water and order (F(1,27) = 7.37, p = 0.011, ηp2 = 0.215), water and hydration (F(1,27) = 7.44, p = 0.013, ηp2 = 0.209), and water and thirst (F(1,27) = 7.69 p = 0.012, ηp2 = 0.212). Participants scored overall higher on the judgment and decision-making tasks in conditions in which they received water compared to the no water day (Fig. 2). There were no simple main effects from factors order, F(1,27) 2.730, p = .110, hydration F(1,27) < 1, or thirst, F(1,27) = 2.33, p = .138. There were also no interaction effects involving order, all Fs < 1. Water supplementation therefore had a positive effect on scores across the battery of judgment and decision-making tasks, relatively independent of levels of thirst and hydration (on the no water day), or order. The ANCOVAs using thirst and osmolality (mean-centred) as co-variates instead of median splitting found the same patterns effects on cognitive reflection scores, with no main or interaction effect of the co-variates (all Fs < 1).
Table 4 Cognitive reflection score means, SDs and F ratios by water condition (water/no water) and post-testing urine osmolality (low/high) as measured on the day participants did not receive any water
This result confirmed hypothesis 2, that water supplementation increased cognitive reflection scores, and this result was not qualified by any interaction.
Correlation analysis
To investigate the possible relationship between hydration variables, judgment scores and executive functions for different water supplementation conditions, we performed correlation analyses. Table 5 shows differing degrees of associations depending on whether the data used was taken from the day participants received water or not.
Table 5 Correlations between cognitive reflection performance scores and water consumption, urine osmolality, thirst, CANTAB tasks and for both days (participants received/did not receive water) (N = 29)
There were significant correlations between cognitive reflection scores and ChoiceRT (water: r = − 0.473, p = 0.010; no water: r = − 0.579, p = 0.001) and IED errors (water: r = − 0.533, p = 0.003; no water: r = − 0.578, p = 0.001) on both days, water and no water, respectively. There was also a significant correlation between CRT scores and RVP errors on the water day only (r = − 0.451, p = 0.014). All correlations were in the predicted direction with better performance (lower errors or shorter RTs) in executive function tasks being associated with higher cognitive reflection performance. Hypothesis 3 was therefore confirmed—performance on executive function tasks (though only in the water condition for RVP) was associated with higher performance in the judgment and decision-making tasks.
Finally, a linear regression analysis was performed using difference scores (water–no water). The difference in cognitive reflection scores between the no water and the water condition served as the dependent variable (criterion) and the difference (between water and no water day) in ChoiceRT and IED errors as independent variables. The regression model tested whether differences (between sessions) in the executive function tasks were associated with differences in the cognitive reflection scores. Recall that the hypothesis was based on the premise by dual process theories that increased inhibition processes are related to increased performance in CRT-like puzzles and heuristic vignettes. Some approaches in the dual systems framework (e.g., Evans & Stanovich, 2013) further implicate mental simulation performance, the ability to maintain and symbolically manipulate separate mental representations of a problem.
ChoiceRT latency difference scores were log-transformed to reduce potential issues of positive skew and normality of residuals. Results of the multiple linear regression indicated that there was a combined significant effect of differences in ChoiceRT and IED (errors) explaining differences in cognitive reflection scores, (F(2,26) = 3.765, p = 0.037, R2 = 0.224). ChoiceRT difference (t = − 2.244, p = 0.034) was a significant predictor in the model, but not IED error difference (t = − 1.543, p = 0.135) (Table 6). Adding RVP errors (difference scores) as a predictor variable again showed a relationship for cognitive reflection scores with ChoiceRT (t = − 2.343, p = 0.027) but not RVP errors (t = − 1.005, p = 0.324), with the overall model marginally significant, (F(2,25) = 4.899, p = 0.058, R2 = 0.254). Thus, hypothesis 4c was retained: cognitive reflection scores for sessions in which water was given were differentially influenced by ChoiceRT scores compared to sessions in which water was not given—the higher the differences in ChoiceRT latencies (and therefore the worse the inhibition performance between no water and water condition), the lower the improvement of cognitive reflection scores from no water to water condition.
Table 6 Summary of regression analysis for variables predicting cognitive reflection score differences (N = 29) between sessions
Hypothesis 4a and 4b were therefore not retained—differences in tasks measuring attention performance (RVP) or mental simulation (IED) between the water supplementation conditions were not associated with the difference in cognitive reflection scores.