Family Bonding with Universities

Abstract

One justification offered for legacy admissions policies at universities is that that they bind entire families to the university. Proponents maintain that these policies have a number of benefits, including increased donations from members of these families. We use a rich set of data from an anonymous selective research institution to investigate which types of family members have the most important effect upon donative behavior. We find that the effects of attendance by members of the younger generation (children, children-in-law, nieces and nephews) are greater than the effects of attendance by the older generations (parents, parents-in-law, aunts and uncles). Previous research has indicated that, in a variety of contexts, men and women differ in their altruistic behavior. However, we find that there are no statistically discernible differences between men and women in the way their donations depends on the alumni status of various types of relatives. Neither does the gender of the various types of relatives who attended the university seem to matter. Thus, for example, the impact of having a son attend the university is no different from the effect of a daughter.

Appendix

Derivation of the Estimating Equations

In this Appendix we derive the equations that comprise our canonical econometric model. As in Cunningham and Cochi-Ficano (2002), we assume a conventional utility-maximization framework. Consider an alumnus whose utility depends on the consumption of two commodities. The first is a composite numeraire consumption good, C, and the second is donations to her alma mater, D. The individual’s problem is to maximize a utility function, U,

$$ {\text{U}} = {\text{U}}\left( {{\text{C}},{\text{D}};{\text{Z}}} \right) $$

subject to

$$ {\text{p}}_{\text{D}} \cdot{\text{D}} + {\text{C}} = {\text{M}}, $$

where Z is a vector of the individuals’ attributes that affect the amount of utility that she derives from her consumption bundle, p_D is the opportunity cost of a dollar of donations, and M is money income. In our context, Z includes, inter alia, variables that characterize the individual’s affinity to her school, including which members of her family attended.¹³ Then the utility maximizing amount of donations, D^*, is given by

$$ {\text{D}}^{ *} = {\frac{{{{ - \partial {\text{V(p}}_{\text{d}} ,{\text{M; Z}})} \mathord{\left/ {\vphantom {{ - \partial {\text{V(p}}_{\text{d}} ,{\text{M;Z}})} {\partial {\text{p}}_{\text{d}} }}} \right. \kern-\nulldelimiterspace} {\partial {\text{p}}_{\text{d}} }}}}{{{{ - \partial {\text{V(p}}_{\text{d}} ,{\text{M;Z}})} \mathord{\left/ {\vphantom {{ - \partial {\text{V(p}}_{\text{d}} ,{\text{M;\;Z}})} {\partial {\text{M}}}}} \right. \kern-\nulldelimiterspace} {\partial {\text{M}}}}}}} $$

where V(•) is the indirect utility function.

As noted in the text, in our data there are a substantial number of observations in which individuals make no donations in a given year. Several statistical models can be used to estimate the model in this situation (See, Ground and Koch 2007) The approach that makes the most sense in our context is the complete dominance hurdle model, which has been used in previous empirical studies of charitable giving (see, for example, Huck and Rasul 2007). In this model, in the first stage, an individual decides whether or not to make a gift. In the second stage, conditional on having made the decision to give, the individual gives a positive amount. Intuitively, because an individual can give as much or as little as she wants in a given year, once she has decided to make a gift, there is no second hurdle to clear when determining the amount. In essence, we assume that anyone who wishes to make a gift does so.

To implement this framework, we follow Florkowski et al. (2000). We define Ω to be the variable that equals one if the individual decides to make a donation in the first stage; otherwise it is zero. Assuming linearity,

$$ \Upomega^{*} = {\text{X}}\theta + \varepsilon $$

where θ is a parameter vector and ε is a normally distributed error term. Ω equals 1 if Ω^* > 0 and 0 otherwise. Again assuming linearity, we can write

$$ {\text{D}}^{*} = {\text{X}}\beta + \mu $$

where beta is a parameter vector, and mu is a random error. Observed donations, D, equal desired donations, D* as long as Ω is greater than zero:

$$ {\text{D}} = {\text{D}}^{*} \,{\text{if}}\, \, \Upomega = 1 $$

If not,

$$ {\text{D}} = 0{\text{ if}}\, \, \Upomega = 0 $$

As demonstrated in Florkowski et al. (2000), assuming that μ and ε are normally distributed and uncorrelated, then the first stage equation is a probit, and the second stage equation can be estimated by ordinary least squares, using only the observations with positive values of donations. This is the statistical approach used in the text.