Raising Charitable Children: The Effects of Verbal Socialization and Role-modeling on Children’s Giving Mark Ottoni-Wilhelm ∗ Ye Zhang David Estell † Neil Perdue ‡ ∗ IU ∗ Lilly Family School of Philanthropy and Department of Economics (IUPUI) † Indiana University Bloomington ‡ University of Indianapolis Spring School on Generosity and Well-Being, University of Warwick - April 9-11, 2013 Motivation I Voluntary contributions to charitable organizations are an important funding source for public goods. Table : Voluntary contributions Canada Netherlands U.K. U.S. Percent who give 64% 78% 72% 65% Share of GDP 0.72% 0.34% 0.73% 1.44% Motivation I Social importance of voluntary giving reflected in government policy that encourages giving through matches, tax deductions, credits. I Despite these incentives, there are powerful self-interested free-riding tendencies that left unchecked would reduce voluntary giving to very low levels (Andreoni 1988). I Warm glow — utility derived from the act of giving to a charity — counters free-riding (Andreoni 1989). I Warm glow must be socialized, starting in childhood. Motivation I But how? I What socialization techniques are effective in raising charitable children? Introduction I Extensive experimental literature in developmental psychology has established the existence of causal effects of role-modeling and verbal socialization on children’s giving in the lab (Eisenberg, Fabes, and Spinrad 2006). I consensus, but . . . I Eisenberg and Mussen (1989): “It cannot be assumed that procedures that prove to be effective in laboratory studies (such as modeling) will necessarily have a significant and lasting impact on behavior when introduced in a natural setting such as the home . . . Thus, it is important to test in natural settings the effectiveness of those procedures that promote prosocial behavior in the laboratory.” Introduction I Nearly 25 years later, there is still little empirical evidence about the magnitude of causal effects of role-modeling and verbal socialization from nationally-representative data collected from children’s home settings. I Cross-sectional regressions find that both parental role-modeling and verbal socialization have large associations with children’s giving money to charities (Brown, Srivastava, and Taylor 2012; Ottoni-Wilhelm, Estell, and Perdue 2011). Parental role-modeling and verbal socialization actions are endogenous, potentially correlated with child-specific unobserved heterogeneity: I I I Prosocial endowments – time-constant Prosocial environments/values – time-varying Parents may adjust role-modeling and verbal socialization to changes in a child’s prosociality I A parent observes her child exogenously becoming more prosocial (increasing prosocial values): I May think that, because the child is becoming more prosocial on his own, she can reduce the parenting effort devoted to socializing the child to give to charity... I I I ... and increase efforts devoted to socializing him to achieve other behavioral goals, ... or increase the effort devoted to socializing his sibling to give to charity. Alternatively, the parent may realize that now is the time, and reinforce the teachable moment by talking more to her child about giving to charity... I ... and showing the child her example of giving to charity. This Paper I I Estimates of the causal effects of verbal socialization and role-modeling that use nationally-representative data from children in their home settings. Individual and sibling fixed effects specifications and two additional specifications seldom considered: I I Across-time sibling fixed effects Difference over time between siblings This Paper I Maps the identifying assumptions necessary to infer causal effects to assumptions about parental: I Compensation versus reinforcement I I Intra-household allocation (Becker and Tomes 1976) Substitution versus complementarity I I Intra-household allocation (Becker and Tomes 1976, 1979) Cultural transmission and socialization (Bisin and Verdier 2011) This Paper - substantive results I The pattern of our estimates indicates that parents socialize a child differently depending upon changes in his/her sibling’s prosocial values. I Our estimates imply that talking to children about giving raises the probability they give by at least 0.13. I Our estimates indicate no evidence that parents’ role-model of giving has a causal effect on children’s giving. Different estimated effects among demographic sub-groups. I I Parental role-modeling affects giving by ... boys or girls (you guess). Three audiences 1. Charitable giving, voluntary contributions – substantive results about socialization of warm glow. 2. Experimentalists – external validity of laboratory results. I I Carpenter, Connolly, and Meyer (2008) Henrich, Heine, and Norenzayan (2010) 3. Children’s outcomes – investment in human capital (e.g., education, health). I I Econometric strategy – indiv. FE, sibling FE + using the intra-household allocation and cultural transmission literatures to establish the identification framework. Generally applicable to estimate causal effects of parental socialization/investment actions on children’s outcomes. Roadmap 1. Econometric Model and Identification Framework 2. Data: Panel Study of Income Dynamics (PSID) + Center on Philanthropy Panel Study (COPPS), Child Development Supplement (CDS) 3. Empirical Results 4. Discussions 5. Conclusion Econometric Model 1 - the child Consider a model of charitable giving Yijt for child i in family j at time t: Yijt = βXijt + µj + αij + θijt + vijt I Xijt : parent’s socialization action I µj : family-specific time-constant prosocial endowment I αij : child-specific prosocial endowment I θijt : time-varying idiosyncratic prosocial values I vijt : random shocks to the child’s giving For a two-child family, sibling k’s outcome equation: Ykjt = βXkjt + µj + αkj + θkjt + vkjt Econometric Model 2 - the parent Parental socialization decisions: Xijt = γ1 αij + γ2 αkj + δµj + π1 θijt + π2 θkjt + uijt Xkjt = γ1 αkj + γ2 αij + δµj + π1 θkjt + π2 θijt + ukjt I uijt and ukjt : random socialization shocks unrelated to children’s endowments and idiosyncratic prosocial values I δ: effect of the family-specific time-constant prosocial endowment on parent’s socialization of both children I γ1 , γ2 : parent’s socialization responses to the child-specific and his sibling’s specific time-constant prosocial endowments I π1 , π2 : parent’s socialization responses to the child’s and his sibling’s idiosyncratic time-varying prosocial values Child and Sibling Fixed Effects I The child fixed effects estimator plim b T = β + I π1 σθ2 (π12 + π22 )σθ2 + σu2 The sibling fixed effects estimator plim b J = β + (γ1 − γ2 )σα2 + (π1 − π2 )σθ2 (γ1 − γ2 )2 σα2 + (π1 − π2 )2 σθ2 + σu2 Two additional sibling estimators, seldom considered I Difference over time between siblings plim b JT = β + I (π1 − π2 )σθ2 (π1 − π2 )2 σθ2 + σu2 Across-time sibling fixed effects estimator plim b JDT = β + (γ1 − γ2 )σα2 + π1 σθ2 (γ1 − γ2 )2 σα2 + (π12 + π22 )σθ2 + σu2 Identification Framework I T JT Our estimates indicate γ\ 1 − γ2 = 0, focus on b and b . plim b T = β + plim b JDT = β + plim b J = β + (γ1 − γ2 )σα2 + π1 σθ2 (γ1 − γ2 )2 σα2 + (π12 + π22 )σθ2 + σu2 (γ1 − γ2 )σα2 + (π1 − π2 )σθ2 (γ1 − γ2 )2 σα2 + (π1 − π2 )2 σθ2 + σu2 plim b JT = β + I π1 σθ2 (π12 + π22 )σθ2 + σu2 (π1 − π2 )σθ2 (π1 − π2 )2 σθ2 + σu2 Therefore we focus on b T and b JT when developing the identification framework. Identification Framework - focus on b T and b JT I We map each (π1 , π2 ) pair to qualitative assumptions about parental: 1. Compensation-versus-reinforcement in response to changes in the difference between children’s prosocial values 2. Substitution-versus-complementarity in response to changes in the sum of children’s prosocial values I We can then map these assumptions to the directions and relative magnitudes of the biases in the b T and b JT estimators Identification Framework (1a) - Intra-household allocation p2 Quadrant 1 C om R ei pe nf ns or at ce e 2 Quadrant 4 1 p1 - p2 < 0 0 p1 om pe C -2 R ei nf or ce ns at e -1 p1 - p2 > 0 Quadrant 3 -2 Quadrant 2 -1 0 1 2 Identification Framework (1b) - Cultural transmission p2 2 Quadrant 4 Quadrant 1 T EN TE TU EM TI PL BS SU M O C 1 p1 + p2 > 0 0 p1 -1 p1 + p2 < 0 C -2 T EN TE TU EM TI BS SU PL M O Quadrant 3 -2 Quadrant 2 -1 0 1 2 Identification Framework (1c) - Cultural transmission + Intra-household allocation p2 Quadrant 1 1 T EN TE TU EM TI PL BS SU M C om R ei pe nf ns or at ce O C e 2 Quadrant 4 -2 ce R ei nf or C T EN TE TU EM TI BS SU om pe PL M ns O at C e -1 0 p1 Quadrant 3 -2 Quadrant 2 -1 0 1 2 Identification Framework (1d) - Cultural transmission + Intra-household allocation I The π2 = π1 and π2 = −π1 lines partition the π1 , π2 plane into four triangles 1. 2. 3. 4. Compensation and SUBSTITUTION (the west triangle) Reinforcement and SUBSTITUTION (the south triangle) Reinforcement and COMPLEMENTARITY (the east triangle) Compensation and COMPLEMENTARITY (the north triangle) Identification Framework (2a) - Directions of bias in b T and b JT I The π1 , π2 plane can be partitioned into regions characterized by different directions of bias in b T and b JT . Identification Framework (2b) - Directions of bias in b T and b JT p2 Quadrant 1 T EN TE TU EM TI PL BS SU M C om R ei pe nf ns or at ce O C e 2 Quadrant 4 T b + JT 1 b - 0 -- p1 -1 + + T at ns pe + JT -2 R ei nf or ce b T EN TE TU EM TI BS SU om PL M O C - C e b Quadrant 3 -2 Quadrant 2 -1 0 1 2 Identification Framework (3a) - Magnitudes of bias in b T and b JT I The π1 , π2 plane can be partitioned into regions in which the magnitude of the bias in either b T or b JT is smaller. Identification Framework (3b) - Magnitudes of bias in b T and b JT p2 Quadrant 1 1 T EN TE TU TI BS SU EM PL M O C C om R ei pe nf n or ce sat e 2 Quadrant 4 0 p1 1/2 1/2 R O C C om ei pe nf n or ce sat e -1 -R 2 -2 Quadrant 3 -2 T EN TE EM ITU T BS SU R 2 PL M R = su / sq Quadrant 2 -1 0 1 2 Identification Framework - Summary p2 Quadrant 1 T 1 T EN TE TU TI BS SU EM PL M O C C om R ei pe nf n or ce sat e 2 Quadrant 4 JT JT T 0 p1 1/2 1/2 -R R T JT C C om ei pe nf n or ce sat e -1 JT T EN TE EM ITU PL ST B SU M O -2 R T Quadrant 3 -2 Quadrant 2 -1 0 1 2 Data sources: PSID, COPPS, and CDS I PSID: Nationally representative longitudinal sample of over 22,000 individuals living in 8,000 families in the United States, 1968ff. I COPPS: Generosity module in the PSID, 2001ff. I CDS: Started in 1997, children from PSID-responding family units. From each family unit up to two children aged 0-12 were selected. CDS–2 in 2002 and CDS–3 in 2007. Data: key variables I CDS–2 and CDS–3 children ages 8 and older were asked “Did you give some of your money last year–if only a few pennies–to a church, synagogue, or another charity that helps people who are not part of your family?” I The child’s primary care giver, usually the child’s mother, was asked “Do you ever talk to (CHILD) about giving some of (his/her) money–if only a few pennies–to a church, synagogue, or another charity?” I Information on giving to religious congregations and charitable organizations from the PSID’s COPPS module 2001-2007 waves form our role-modeling variables. Data: samples 1. Child fixed effects sample: 936 children who were old enough in CDS–2 to have been asked the giving question, and young enough in CDS–3 to have not aged out. I 144 pairs in which both siblings are present in CDS–2 and CDS–3. We use this sample in our difference over time between siblings model. 2. Sibling fixed effects sample: 576 sibling pairs (1,152 children) in which both children were old enough in CDS–2 to have been asked the giving question. 3. Across-time sibling fixed effects sample: 521 sibling pairs (1,042 children) in which older sibling in CDS–2 and a younger sibling in CDS–3. Descriptive Statistics 1a - Child giving over time 0 .15 Fraction .3 .45 .6 Giving, same child, over time 0 -> 0 1 -> 0 0 -> 1 1 -> 1 Descriptive Statistics 1b - Parent talking over time 0 .15 Fraction .3 .45 .6 Parent talking, same child, over time 0 -> 0 1 -> 0 0 -> 1 1 -> 1 Descriptive Statistics 1c - Parent role-modeling over time 0 .15 Fraction .3 .45 .6 Parent role-model, same child, over time 0 -> 0 1 -> 0 0 -> 1 1 -> 1 Descriptive Statistics 2a - Giving: younger sib, older sib 0 .15 Fraction .3 .45 .6 Giving, sibling pair: younger (first), older (second) 0,0 1,0 0,1 1,1 Descriptive Statistics 2b - Parent talking: younger, older 0 .15 Fraction .3 .45 .6 Parent talking, sibling pair: younger (first), older (second) 0,0 1,0 0,1 1,1 OLS vs. Child Fixed Effects Estimates Table 3. Cross-sectional and Within-child Estimates of Effects of Verbal Socialization and Role-modeling OLS Parent talks about giving Role-modeling: parent gives Observations 0.135*** (0.025) 0.062*** (0.022) 1,858 Dependent Variable: Child's Giving Child FE 0.119*** 0.131*** 0.128*** (0.026) (0.036) (0.037) 0.064** -0.005 -0.004 (0.025) (0.037) (0.038) 1,839 1,806 1,806 Source: Authors' calculations from PSID and CDS samples. Notes: 1. Different numbers of observations for columns are due to missing values on some of the independent variables, results do not change if all the regressions are based on the smallest sample. 2. Robust standard errors clusted at individual level are in parentheses. 3. *** p<0.01, ** p<0.05, * p<0.1 4. All regressions include a year dummy. Sibling Fixed Effects Estimates Table 4. Different Types of Within-sibling Estimates of the Effects of Verbal Socialization and Role-modeling Sibling FE Parent talks about giving Dependent Variable: Child's Giving Difference Over Time Across-time Sibling FE Between Siblings 0.027 (0.074) 0.007 (0.070) -0.067 (0.149) -0.120 (0.147) 0.142*** (0.048) -0.068 (0.056) 0.125** (0.050) -0.061 (0.055) 1,152 1,152 288 288 1,042 1,042 Role-modeling: parent gives Observations Source: Authors' calculations from PSID and CDS samples. Notes: 1. Robust standard errors are in parentheses. 2. *** p<0.01, ** p<0.05, * p<0.1 Discussion - no identifying assumptions required I Parents neither compensate nor reinforce differences between their children’s prosocial endowments, that is γ1 = γ2 . I I Parents do socialize child i differently depending upon time-varying changes in child k’s prosocial values, that is π2 6= 0. I I ˆ and bˆJ ≈ bˆJT bˆT ≈ b JDT bˆT 6= bˆJT Some π1 , π2 combinations can be ruled out (e.g., substitution combined with reinforcement) Identification Framework (again) I T JT Our estimates indicate γ\ 1 − γ2 = 0, focus on b and b . plim b T = β + plim b JDT = β + plim b J = β + (γ1 − γ2 )σα2 + π1 σθ2 (γ1 − γ2 )2 σα2 + (π12 + π22 )σθ2 + σu2 (γ1 − γ2 )σα2 + (π1 − π2 )σθ2 (γ1 − γ2 )2 σα2 + (π1 − π2 )2 σθ2 + σu2 plim b JT = β + I π1 σθ2 (π12 + π22 )σθ2 + σu2 (π1 − π2 )σθ2 (π1 − π2 )2 σθ2 + σu2 Therefore we focus on b T and b JT when developing the identification framework. Discussion - π1 , π2 combinations ruled out p2 Quadrant 1 T 1 T EN TE TU TI BS SU EM PL M O C C om R ei pe nf n or ce sat e 2 Quadrant 4 JT JT T 0 p1 1/2 1/2 -R R T JT C C om ei pe nf n or ce sat e -1 JT T EN TE EM ITU PL ST B SU M O -2 R T Quadrant 3 -2 Quadrant 2 -1 0 1 2 Discussion - Identifying assumption of cultural substitution I I Child-fixed estimate (.13) is a lower bound to the causal effect of verbal socialization. Two supporting reasons for cultural substitution 1. If complement then (a) Intra-household literature indicates that some other use of parental time must be a non-normal good, or (b) Some other theory of parental socialization must be used (e.g., bounded rationality) 2. If complement then (a) Cultural transmission literature predicts that giving should be homogeneous in the population - everyone should be giving to charity, but (b) The evidence is that giving is heterogeneous in the population (between 64% - 78% give) Discussion - Cultural substitution (continued) I An additional check on the identifying assumption of substitution is that substitution + the pattern of estimates we get rules out parental reinforcement of within-sibling changes in their childrens prosocial values. It would then follow that parents must be must be compensating these changes. I I Compensation implies parents adjust their socialization actions in response to the specific needs of their children. An influential view in child development is that best parenting practices do exactly that. Cultural substitution is sufficient, but not necessary, for the .13 to be a lower bound. I I Necessary condition is weak complementarity. To argue for a zero causal effect requires an identifying assumption of strong complementarity ... I I plus very weak (zero) parental compensation of within-sibling changes in prosocial values, or reinforcement. Discussion - summary figure (if needed) p2 Quadrant 1 T 1 T EN TE TU TI BS SU EM PL M O C C om R ei pe nf n or ce sat e 2 Quadrant 4 JT JT T 0 p1 1/2 1/2 -R R T JT C C om ei pe nf n or ce sat e -1 JT T EN TE EM ITU PL ST B SU M O -2 R T Quadrant 3 -2 Quadrant 2 -1 0 1 2 Discussion - Role-modeling I The child fixed effects estimates taken at face value imply a zero causal effect of role-modeling. I Demographic differences I I Large role-modeling effect among non-African-American girls (.14). Large effects of parents talking about giving among African-American girls and non-African-American boys (.18, .22). Conclusions I Provides estimates of the causal effects of verbal socialization and role-modeling on children’s charitable giving based on nationally-representative data from children in their home settings. I We develop a framework useful for identifying causal effects of parental socialization/investment actions on children’s outcomes in the presence of time-varying unobserved heterogeneity. Under the identifying assumption we think is most reasonable—that parents treat the socialization actions of others as cultural substitutes—our estimates imply that talking to children about giving raises the probability they give by at least 0.13. I I Parents whose goal is to socialize warm glow should continue, not curtail, conversations about giving as their children age through adolescence. Conclusions I Surprisingly, in light of the developmental psychology experiments, our estimates based on the full sample of children indicate no evidence that parents’ role-model of giving has a causal effect on children’s giving. I I I Laboratory results on role-modeling were not externally validated, except for non-Afr. Amer. girls. Translational research is needed to learn how role-modeling, known to be effective in the laboratory, can be used effectively in the home. The effects of verbal socialization and role-modeling vary across demographic groups. Thank you! I Centre for Competitive Advantage in the Global Economy I Centre for Market and Public Organisation I Economic and Social Research Council I Department of Economics, University of Warwick I Notre Dame Science of Generosity Project Foundations that fund the collection of generosity data in the PSID I I Atlantic, Gates, Fidelity Charitable, John Templeton, Mott, Corporation for National and Community Service, and others like you. Child Fixed Effects Estimates: Demographic Sub-groups Table 6. Within-child Estimates of Effects of Verbal Socialization and Role-modeling for Different Demographic Groups Dependent Variable: Child's Giving African-Amer. non-African-Amer. African-Amer. non-African-Amer. girls girls boys boys Parent talks about giving 0.184** 0.071 0.081 0.216*** (0.087) (0.066) (0.080) (0.067) Role-modeling: parent gives -0.135 0.143* 0.065 -0.073 (0.087) (0.074) (0.071) (0.069) Observations 376 478 420 532 Source: Authors' calculations from PSID and CDS samples. Notes: 1. Robust standard errors clusted at individual level are in parentheses. 2. *** p<0.01, ** p<0.05, * p<0.1 3. All regressions include a year dummy.