Results

Behavioral parameters
The average amount sent from the first player (Trust) is 11.1 out of 20, while the average amount sent back to the second player (Trustworthiness) is 10.0 (Figure S1A, Figure S1B). This is similar to what is generally observed in the TG literature: the first player sends about 50% of his/her endowment and the second player sends back about 95% of what was sent (see [24] for a survey). In a linear regression, males significantly trust more than females (p<0.039), while we observe no significant gender difference in trustworthiness (p>0.363) (Figure S2A and Figure S2B). Our results are distinct from [35] which reported significant gender difference in trustworthiness but not in trust. This discrepancy likely indicates slight cultural differences across populations.

Plasma OT levels
The average oxytocin level is 214± SD 230 pg/ml (Figure S1C) which is representative of the assay procedure used in the current investigation [8]. We test whether age and gender might have an effect on plasma OT level. As usually observed [36], [37], in linear regression, there is no significant gender difference in plasma OT levels (p>0.365) (Figure S2C), and also no significant age effect on Plasma OT (p>0.650). Following other investigations [38], [39], we exclude 23 subjects whose plasma OT exceed 3 standard deviations (>904) from the subsequent analysis (Figure S1C), and then use the log transformation in the analysis.

Relationship between plasma OT, Trust and Trustworthiness
We test the effect of plasma OT on the levels of trust and trustworthiness in the TG using both linear and non-linear regression analysis; the results are summarized in Tables 1 and 2. Neither Trust nor Trustworthiness shows a significant linear relationship with plasma OT (Table 1, Table 2) whereas a significant non-linear U-shaped (quadratic) relationship is observed with Trust (OT, p<0.001; OT quadratic, p<0.002) (Figure 1A, 1B). Subjects in the top 20% and the bottom 20% of the plasma OT distribution “trust” on the average 15.6% more than those subjects in the middle 20% of the distribution (Figure 1B). Hence, subjects characterized at the extremes of plasma OT concentrations are significantly more trusting. After separating the analysis by sex, the significant relationship is observed only in male subjects (oxytocin: p<0.002; oxytocin quadratic: p<0.002), but not in female subjects (OT: p<0.232; OT quadratic: p<0.250) albeit in the joint analyses greater significance is observed.
10.1371/journal.pone.0051095.t002 Table 2  Regression results for linear and nonlinear relationship between plasma oxytocin and trustworthiness.     The table reports coefficient, standard error, and p values. The last row is R-squared.   For average Trustworthiness, we find a significant U-shaped relationship between plasma OT and Trust (OT, p<0.041; OT quadratic, p<0.043) (Figure 2A, 2B) similar to that observed for Trust. Subjects in the top 20% and bottom 20% of the plasma plasma OT distribution are 8.3% more trustworthy than those in the middle 20% plasma OT distribution (Figure 2B). In the analysis separating by sex, similar to what we observed for Trust, a marginally significant relationship is again observed only in the male subjects (OT: p<0.065; OT quadratic: p<0.065), but not in female subjects (OT: p<0.210; OT quadratic: p<0.240).
10.1371/journal.pone.0051095.g002 Figure 2  Plasma oxytocin and trustworthiness.
(A) Scatter Plot on the relationship between plasma oxytocin and trustworthiness. (B) Histogram on the relationship between plasma oxytocin and trustworthiness.   We check the robustness of the results after including those subjects with plasma OT higher than 3 times of standard deviations. Similarly, a significant non-linear U-shaped relationship is observed with Trust (OT, p<0.024; OT quadratic, p<0.028), and a marginally significant non-linear U-shaped relationship is observed with Trustworthiness (OT, p<0.071; OT quadratic, p<0.070). We further check the robustness of the results after controlling gender and age in the regression analysis. Similarly, a significant non-linear U-shaped relationship is observed with Trust (OT, p<0.002; OT quadratic, p<0.002), and a significant non-linear U-shaped relationship is observed with Trustworthiness (OT, p<0.026; OT quadratic, p<0.029).
We also consider whether there might be a relationship between the decision to send and return. In our data, 11.5% of the subjects send nothing, and 10.2% of the subjects return nothing. However, in the probit regression, we do not observe a significant relationship between plasma oxytocin and the decision to send (OT, p>0.839; OT quadratic, p>0.832), nor the decision to return (OT, p>0.892; OT quadratic, p>0.895). This might be due to lack of power as the data is skewed for the binary decisions. Alternatively, it might suggest that the plasma OT is more reflecting the degree of trust and trustworthiness, rather than per se trust or not, and trustworthy or not.
Trust by the first mover is risky since the trustee may not reciprocate the amount transferred. As anticipated, risk attitude and trust are significantly correlated (corr  = 0.141, p<0.001) and the direction is positive viz., subjects with a propensity to financial risk are more trusting in the TG. Towards testing the specificity of plasma OT levels as a measure of the level of trust rather than degree of risk tolerance, we conduct a regression analysis with risk attitude as an additional independent variable. The inverted U-shaped association between trust and plasma OT remains robust (the whole sample: OT, p<0.001; OT quadratic, p<0.001; male only: OT, p<0.002; OT quadratic, p<0.002; female: OT, p<0.121; OT quadratic, p<0.133). We further test the relationship between risk and OT; no significant relationship is found (the whole sample: OT, p<0.693; OT quadratic, p<0.594; male only: OT, p<0.695; OT quadratic, p<0.812; female: OT, p<0.247; OT quadratic, p<0.264). This further supports the view that the effect of OT is specifically on trust but not risk attitude.