Parenting is risky business: parental risk attitudes in small stakes decisions on behalf of their children

Abstract

Parents often face risk when making decisions on behalf of their children, since outcomes may affect child development. We perform an incentive-compatible field experiment using the Holt and Laury (Am Econ Rev 92(5):1644–1655, 2002) design to elicit parental risk preferences in a stewardship decision framework. Multivariate analysis using different estimation techniques suggests that parents are significantly more risk-averse when deciding for their child than for themselves. Higher risk aversion is linked to characteristics of parents, not of children. Mood and gender of the deciding parent play a key role. If these results also hold for larger stakes, insights from this study could help to improve decision environments for parents to limit inequality between children due to diverging parental risk preferences.

Appendices

Appendix 1: Comments on the sample

The recruitment of participating parents and the conduct of the experiments are quite time-consuming. To reduce time and effort we chose larger kindergartens to address parents in the relevant circumstances. Since kindergartens are often small (often only 1 or 2 groups) and due to logistical reasons, we concentrated on kindergartens with at least 50 children in the southern region of Hannover, which is the capital of Lower Saxony in Germany. Therefore, our parental sample is from a rather rural population and our sample does not capture typical social problem mixtures that are more common in city centers.

Furthermore, we are only able to provide instructions and communication in German which is a barrier for foreigners to participate. Hence, parents and children with a migration background are under-sampled. Moreover, participation was voluntary and only interested parents took part. Since time is a scarce resource, our final payment for the parents was the driving factor for participation in only very few cases. Instead, we invented the “donation sandglass”. This was a device for convincing the parents to participate without influencing their experimental payment afterwards. We announced that the kindergarten would get a donation of 5.00 EUR from the university for each participant. Participants then got a colored paper like a little tile that he or she glued on the “donation sandglass”. We prepared “donation sandglasses” for every kindergarten and put them on each entrance door. Parents were then able to see how much money the kindergarten had already gained. The “donation sandglass” motivated other parents to participate. The sample selection problem of our sample should not be worse compared to many student samples used in experiments.

For all these reasons our sample is not a representative one. However, there is still some variety in this otherwise relatively homogenous sample, and we obtain interesting results. We conclude from this that the point estimates of our results should not be taken to be representative of the German population but that the general direction of the effects can be interpreted.

Appendix 2: Ordered logit estimation results

Since all regressors are case specific, we estimate an ordered logit model to explain the influence of parent and child characteristics on the number of safer choices to show the robustness of the OLS estimates of Table 7.

$$ \Pr \left( {outcome_{i} = j} \right) = \Pr \left( {\tau_{j - 1} < \beta_{pp} p_{p;i} + \beta_{pt} t_{p;i} + \beta_{cp} p_{c;i} + \beta_{ct} t_{c;i} + \beta_{cc} c_{c;i} + \beta_{d} d_{i} + \varepsilon_{i} < \tau_{j} } \right) $$

where p and c are independent variables containing child and parent characteristics (see description in Subsect. 3.3), t captures the treatment order and ε _j is logistically distributed. We have 10 different outcomes (number of safer choices) resulting in nine cutpoints τ ₁, τ ₂, … τ ₉. These cutpoints are estimated together with the coefficients β. The estimation results in columns (1–7) display odds ratios, standard errors (in parentheses) and their significance level (asterisks). We calculate robust standard errors, which are clustered over parents’ identifiers. The specification of each column is exactly the same as the specifications in Subsect. 3.3 for the OLS estimations.

The decision dummy has a positive influence on the number of safer choices in column 4 and 5 (Table 10). Neither the age of the child, nor his or her gender, nor the existence of siblings has any significant effect on the parental choices. As for the OLS regressions, parent characteristics can explain the higher number of safer choices in the stewardship decision. Even so the gender of the parent is not for all specifications above the 10 % level with respect to the stewardship decision, it is very close to the 10 % level. The same applies to the CRT-value in the individual decision. All other results are confirmed.

Table 10 Estimation results, dependent variable is a dummy, safer choice yes=1/no=0