James Konow

1
Mixed FeelingsTheories and Evidence of Warm
Glow and Altruism

• James Konow
• presentation to
• The Nature and Development of Moral Norms
  Workshop
• CSMN, University of Oslo

2
Two Questions about Altruism

• Intrinsic motive for altruism
• Cognitive based on knowledge, thought or reason
• Affective based on moods, feelings or emotions
• Subject of much empirical and theoretical
  research
• Economics mostly inferred from behavior
• Psychology mostly studied through self-reports
• Extrinsic purpose of altruism
• Conditional altruism conditioned on moral norms
• Unconditional altruism not conditioned on any
  norms, although often implicitly depends on
  personal relationships

3
The sentiment or affection of the heart from
which any action proceeds may be considered
under two different aspects, or in two different
relations first, in relation to the cause which
excites it, or the motive which gives occasion to
it and secondly, in relation to the end which it
proposes, or the effect which it tends to
produce.   Adam Smith, The Theory
of Moral Sentiments
4
Goals of the current study

• Construct simple models of altruism and generate
  behavioral predictions from those models
• Employ a parallel experimental design to
  discriminate behavior predicted by the alternate
  theories
• Collect data on short run (or transient) feelings
  and relate to altruistic behavior first
  economics experiment to do so
• Formulate a theory of conditional altruism to
  reconcile results inconsistent with theories of
  unconditional altruism

5
Unconditional Altruism

• Economic theories of unconditional altruism
• Pure altruism (Becker 1974)
• Preference for anothers material or psychic
  well-being
• Cognitive, unselfish concern for others
• Warm glow (Andreoni 1989, Harbaugh 1998)
• Good feeling from giving
• Selfish concern for ones own happiness
• Impure altruism (Andreoni 1989, 1990)
• Combination of pure altruism and warm glow
• Psychological literature on altruism
• Cognitive (Cialdini) Identification and merging
  with others, so selfishly help others as one
  would oneself
• Affective (Batson) Selfless empathy with those
  in need

6
Behavioral Implications Crowding Out

• Crowding out
• Do increased contributions to a public good
  (e.g., charity) by others reduce (or crowd out) a
  donors own transfers?
• Pure altruism
• Individual since the donor only cares about the
  total benefit to a recipient, a one dollar
  increase in other contributions will reduce his
  voluntary donations by one dollar
• Public policy public spending on public goods
  will crowd out private spending dollar-for-dollar
  (crowding out is complete)
• Impure altruism
• Individual a one dollar increase in other
  contributions will reduce the donors giving but
  by less than one dollar, because he takes
  pleasure in giving per se
• Public policy one dollar in public spending will
  reduce private donations by less than one dollar
  (crowding out is incomplete)

7
Evidence on Crowding Out

• Crowding out appears to be incomplete
• Field studies of actual charitable giving (e.g.,
  Kingma 1989, Payne 1998)
• Public goods experiments (Andreoni 1993, Palfrey
  and Prisbey 1996, 1997) subjects voluntarily
  contribute to a fund shared by all
• Limitations of these studies
• Imperfect information about public spending
• Selfish concern for status or prestige
• Results from public goods experiments potentially
  confounded by subject confusion, expectations or
  strategic considerations

8
Affective Motivation

• Evidence
• Mere possibility of recipients expressing
  emotions about level of donor transfers is
  sufficient to increase transfers (Ellingsen and
  Johannesson, 2008, Xiao and Houser, 2007)
• Neural evidence is consistent with both pure
  altruism and warm glow, i.e., impure altruism
  (Harbaugh et al., 2007)
• Theory
• Few studies present formal models linked to
  empirical measures of happiness, or subjective
  well-being as it is called in the psychology
  literature (Graham and Oswald, 2008, McBride,
  2008)

9
Description ofBasic Experimental Design

• Dictator Game (or Dictator Experiment)
• One subject, called the dictator, is endowed
  with a fixed sum of money
• The dictator may share his/her endowment with an
  anonymous counterpart, called the recipient,
  who is in another room
• Recipients have no recourse and must simply
  accept whatever dictators give (even nothing),
  eliminating strategic concerns on the part of
  dictators

10

• Simple design minimizes subject confusion
  potentially present in more complex experiments,
  e.g., public goods experiments
• All subjects complete a questionnaire asking
  about mood just before and just after the
  allocation decision
• Double blind procedures no one, including the
  experimenter, can associate decisions or
  responses with specific subjects, minimizing
  self-interested reasons for giving (e.g., status,
  prestige) and for misrepresenting true mood
  (e.g., social desirability)

11

• Previous Dictator Findings
• Tax experiment
• (Bolton and Katok 1998, Eckel, Grossman and
  Johnston 2005)
• Two treatments in between subjects design with
  differing endowments of dictators and recipients
• Treatment Dictator Recipient
• Tax 18/2 18 2
• Tax 15/5 15 5
• Tax experiment because it is as if the dictator
  were subject to an involuntary tax that is given
  to the recipient
• Every subject participates in only one treatment
  as dictator
• After decisions, one of the two subjects in a
  pair is randomly chosen to be dictator for actual
  allocation (strategy method)
• Predictions of the model are summarized in
  Proposition 1

12

• Proposition 1
• Under unconditional altruism, crowding out in the
  Tax experiment is partial or complete
• Pure altruism Dictators care only about the
  final distribution and not whether their gifts
  are voluntary or involuntary, thus, crowding out
  is complete
• The more heavily taxed dictators give a full 3
  less in the 15/5 treatment compared to the 18/2
  treatment
• Warm-glow Dictators value the gift itself, and
  crowding out is partial (i.e., incomplete)
• Dictators give less in the 15/5 treatment than
  in the 18/2 treatment but this difference is less
  than 3
• Impure altruism Crowding out is incomplete and
  intermediate to pure altruism and warm glow
• Pure altruism motivates a larger reduction than
  with warm glow alone, but warm glow prevents this
  from being complete

13
Results and Limitations

• Results
• Dictators give less in the 15/15 treatment than
  the 18/2 treatment
• This difference is less than 3, indicative of
  incomplete crowding out
• Limitations of design
• This design does not allow one to differentiate
  warm glow and impure altruism incomplete
  crowding out is consistent with both hypotheses
• Subjects roles having subjects play both roles
  differs from much real world giving and might
  affect the level of transfers
• Strategy method some studies suggest decisions
  differ when each decision is actually paid out,
  especially when social preferences are at issue

14
New Experiments

• Subjects are always in only one role, dictator or
  recipient
• All decisions are actually paid
• Dictators are always student subjects with 10
  endowments
• Treatments differ with respect to recipient
  identity, recipient endowments or possible
  transfers
• Subsidy Experiment
• Recipients are all fellow students whose initial
  endowments vary across treatments
• All endowments are common knowledge
• Two treatments
• Standard treatment Recipients are unendowed
• Subsidy treatment Recipients are endowed with 4
  (like subsidizing dictator gifts to recipients)

15
Predictions for Subsidy Experiment

• Proposition 2
• With unconditional altruism, crowding out in the
  Subsidy experiment is partial or zero
• Warm glow Crowding out is zero, since transfers
  are motivated solely by the act of giving
• Dictator transfers are the same in the Standard
  and Subsidy treatments
• Pure altruism Crowding out occurs, since
  dictators care about the recipients total
  allocation and, therefore, are lower in the
  Subsidy treatment than in the Standard treatment
• Crowding out, however, is incomplete the
  recipients endowment reduces the marginal
  benefit of giving to him
• Impure altruism Crowding out is partial but
  smaller than in the case of pure altruism

16

• Charity experiment
• Recipients are always unendowed
• Treatments differ with respect to whether
  recipients are students or charitable
  organizations
• Two treatments
• Charity treatment Recipients are one of two
  obscure charities, Childreach or Children
  International, one of which each dictator must
  choose if he/she gives anything
• Standard treatment Recipients are fellow
  students whom subjects have seen at start of
  experiment

17

• Eckel and Grossman (1996) found donations to Red
  Cross significantly greater than those to
  anonymous students and conclude
• Familiarity Hypothesis
• Donors are more generous toward recipients about
  whom they have more information, ceteris paribus.
• The current experiment reverses the information
  conditions such that average gifts to the
  relatively unknown charities in the Charity
  treatment should be less than those to known
  students in the Standard treatment

18

• Preference-based crowding out is
• ambiguous in Charity experiment
• If experimental frame takes, there is no
  obvious preference-based reason for dictator
  transfers to differ across treatments
• If dictators interject own assumptions about the
  recipients endowments external to experiment,
  however, transfers might be lower in Standard
  treatment than Charity treatment (although this
  difference is also consistent with an alternate
  hypothesis discussed later)

19

• Matching grant charity manipulation
• On about one-half of forms for subjects in the
  Charity treatment, a statement is added that
  indicates that donating 1 to Childreach will
  make available an additional 3 through a
  matching grant program.
• In other words, the benefit to the recipient for
  every dollar given, or matching grant multiplier,
  equals 4.

20

• Predictions regarding matching grant
• If warm glow is the sole motive, there should be
  no difference in the percentage of subjects
  choosing Childreach over Children International
  because of a matching grant.
• Any shift, however, toward Childreach in the
  version with the matching grant indicates some
  concern for the well-being of recipients,
  consistent with pure or impure altruism.

21

• Proposition 3
• An unconditionally altruistic donors gift to a
  charity is nondecreasing in that charitys
  matching grant multiplier. Specifically, the
  effect on giving of a matching grant multiplier
  of 4 is
• Warm glow zero
• Pure altruism positive
• Impure altruism positive but less than with pure
  altruism
• Conclusion A matching grant increases giving
  with pure or impure altruism but has no effect on
  giving if altruism is motivated solely by warm
  glow.

22
Affective Evidence onWarm-glow

• Measures of changes in mood
• Identify self-reported changes in mood, or short
  run affect (SRA), of dictators by comparing
  measure of mood just after the allocation
  decision with mood just before decision (SRAD).
• Compare the mood changes of more generous
  dictators in each treatment with those of two
  other groups
• Less generous dictators in the corresponding
  treatment (Standard, Subsidy or Charity)
• Control treatment Dictators are paired with
  unendowed student recipients but are prevented
  from transferring any of their endowment the
  endowments and procedures are common knowledge.

23
Summary of Experimental Design

• Four treatments differ as follows
• Standard Recipients are unendowed students 47
  pairs of dictators and recipients
• Subsidy Recipients are students endowed with 4
  58 pairs of dictators and recipients
• Charity Recipients are one of two unendowed
  charitable organizations 47 dictators plus 24
  from pilot study sometimes used
• Control Recipients are unendowed students and
  they remain so dictators have no opportunity to
  give any of their 10 endowment to recipients 45
  pairs of dictators and recipients

24
Experimental Procedures

• Recruitment subjects were required to
  participate in an experiment to satisfy course
  requirement addresses possible self-selection
  bias and generates more variance
• Show-up fee of 5 plus salient earnings
• After being paid, 96 indicated they would be
  willing to participate again in economics
  experiment
• Initially show up to common room and then
  dictators and recipients randomly assigned to
  separate rooms
• Double blind procedures no one can associate
  decisions with a specific subject, including the
  experimenter
• Short run affect score (SRA) derived from
  responses on nine-point scales to good mood-bad
  mood and depressed-elated item change in SRA
  (SRAD) equals SRA after the allocation minus SRA
  before the allocation decision (or announcement
  of payment in Charity treatment)

25

• Sequence of steps
• All subjects complete a questionnaire that
  includes short run affect questions.
• Each dictator then chooses how many of their ten
  1 bills to pass along to a recipient.
• Everyone completes another questionnaire that
  repeats the short run affect questions.
• All forms and gifts are put in a single packet.
• Later follow-up questions ask anonymous
  demographic questions that cannot be associated
  with the earlier decisions.
• Charity treatment is the same except
• Dictators choose to give to one of two charities
  or to neither from brief descriptions.
• Follow-up questions also ask whether they had
  previously heard of these charities.

26
Notes on Deviations from Benchmark Cases

• Subsidy A few dictators in these sessions did
  not properly process the 4 gift based on the
  post-experimental questionnaire, so we also
  consider a Subsidy Subset that excludes six
  confused subjects.
• Charity The four basic Charity sessions used
  same procedures as other treatments two Charity
  pilot sessions differed (e.g., they excluded
  short run affect measures), but allocations in
  the pilots do not differ significantly from basic
  charity sessions, so they are also reported in
  those instances as Pooled Charity.

27
Results of Allocation Decisions

• Gifts are most generous in the Charity treatment,
  followed by Standard and then Subsidy.
• Modal gifts of Givers create equal splits in all
  cases save Charity Standard 5, Subsidy 3 (also
  in earlier Tax 15/5 5 and Tax 18/2 8, not shown
  here).

28

• Formally test crowding out using three parametric
  and nonparametric tests of differences in giving
  (differences in means, Mann-Whitney and
  Kolmogorov-Smirnov)
• According to most tests, crowding out occurs but
  is incomplete (Table 4)
• Giving is significantly greater in the Charity
  treatment than in the Standard treatment,
  contrary to the Familiarity Hypothesis

29

• The results of the Charity experiment cannot be
  attributed to differences in the familiarity of
  subjects with recipients, since mean gifts do not
  differ based on level of familiarity

30
Summary of Matching Grant Results(Table 6)

• Results on choice of charity
• The proportion of dictators choosing the matching
  grant charity counts against warm glow and for
  pure or impure altruism.
• In the normal version in which the grant is not
  mentioned, subjects choose Childreach and
  Children International in about the same
  proportion, but Childreach with the matching
  grant is chosen significantly more often than
  Childreach without the grant or than Children
  International.
• Results on level of giving
• Evidence on the level of giving, however, is
  inconsistent with pure or impure altruism.
• Dictators should give more to Childreach with the
  matching grant, but they do not indeed they give
  less significantly so in one case.

31
Results on Affective Motivation

• Sample in each treatment is bifurcated into
  dictators whose gifts are at or above that of the
  median Giver, or High gifts, and those below
  this, or Low gifts (Table 5)
• Compare changes in mood (SRAD)
• Standard and Subsidy treatments
• Those who give less are significantly happier
  (have a bigger boost in SRA after allocation
  decisions) than those who give more
• Charity treatment
• Dictators who give more are somewhat happier,
  although this is marginally significant
• Control
• Dictators in Control are significantly happier
  after being told they will keep all than before
  being told
• Relative to Control, dictators are less happier
  who give more in the Subsidy treatment and less
  in the Charity treatment (Standard not
  significant)

32
Regressions Results on SRAD

• OLS regression of SRAD on Gift
• SRAD a ßGift e
• These regressions substantiate results with means
• Weak evidence of warm glow in the Charity
  treatment for the normal case but not for the
  matching grant case

33
Warm Glow or Hot Sting?
34
Conclusions about Warm Glow

• Reject warm glow in the Standard and Subsidy
  treatments at statistically significant levels
• People have mixed feelings generosity has an
  unfavorable effect on short run feelings when
  ones recipients are students, but a possibly
  favorable one when they are charities
• These effects are not due to self-selection, as
  evidenced by comparison to the Control Chosen
  levels of giving in the other treatments can at
  best match the SRAD in the Control treatment

35
Conclusions aboutUnconditional Altruism

• The results on feelings add to the previously
  noted findings on allocation decisions that often
  conflict with warm glow, pure altruism and/or
  impure altruism
• Collectively, there are numerous inconsistencies
  in the evidence on allocation decisions and
  feelings with the various theories of
  unconditional altruism

36
Conditional Altruism

• Theory of conditional altruism is an alternate
  approach that reconciles evidence of allocation
  decisions and feelings inconsistent with theories
  of unconditional altruism
• Introduces moral norms as in more general social
  preference literature, which includes
  considerations of fairness, need, efficiency,
  reciprocity, etc.
• Moral norms depend on context (e.g., Bicchieri,
  2006) here employ context dependent distributive
  norms
• Happiness is also related to different norms that
  depend on the context (e.g., Railton, 2006) here
  focus on short run happiness

37
Model of Conditional Altruism

• Conditional altruism incorporates a simple, but
  important, consideration in decisions about
  giving a condition (f) that a donor believes to
  be the right gift to the recipient
• To be more exact, the condition is the right
  benefit that the gift should produce (the two are
  the same in this study, except in the case of the
  matching grant charity where the benefit
  generated is four times the gift)
• Utility (preferences) are a function of ones own
  allocation (material utility), conditional
  altruism that relates the gifts benefit to f
  (like inequity aversion) and a term that depends
  solely on giving (like warm glow)

38
Conditional Altruism Term

• xj is the gift
• fj is the condition
• f() is the conditional altruism preference term
  of the utility function

39
Determination of the Condition (f)

• In the context of this study, the two most
  salient distributive standards are
• Equity
• In contextually simple studies like this one,
  equity corresponds to equal splits
• Equity applies in Standard, Tax and Subsidy
  treatments with student cohorts
• Need
• This principle calls for satisfaction of basic
  needs and requires a larger transfer than equity,
  when basic needs are not satisfied with equal
  splits
• Need applies in Charity treatment with
  organizations that serve needy recipients

40

• Stylized facts from numerous previous studies
• Dictators who are stakeholders exhibit, on
  average, a self-centered bias in their
  allocations and in their beliefs about right
  allocations in comparison to impartial third
  party spectators who have no stakes
• Nevertheless, a majority of stakeholder dictators
  who give anything at all (the Givers) allocate
  the same impartial amount chosen by spectators
  this permits one to infer the impartial f from
  the modal decisions of stakeholders
• The formal analysis is qualitatively the same
  using the impartial standard or the biased belief
  for f, but acknowledging possible differences
  between impartial gifts and biased beliefs help
  explain certain patterns in the results later

41
Predictions of Conditional Altruism

• Proposition 4
• Assuming conditional altruism, crowding out in
  the Tax experiment is partial
• This is similar to the case of impure altruism
• Proposition 5
• With conditional altruism, crowding out in the
  Subsidy experiment is partial, specifically, less
  than one-half
• Together, these propositions are more specific
  than unconditional altruism and also imply that
  crowding out in the Tax experiment is less than
  one-half that in the Subsidy experiment

42

• Proposition 6
• A conditionally altruistic donors gift changes
  in direct relationship to, but by less than, any
  change in the amount the donor believes to be
  right, i.e., if the condition rises by 1, the
  donor will increase his gift but by less than 1
• Comparing the Charity and Standard treatments,
  this means dictator gifts will be greater in the
  former versus the latter treatment, but by less
  than the difference in f
• This would be true, even if preferences did not
  depend on the size of the gift per se

43

• Proposition 7
• Under conditional altruism, the effect of a
  matching grant program on gifts to that program
  is ambiguous. A sufficient condition for the
  optimal gift with matching grant to be lower,
  however, is that the optimal gift in the normal
  case without the matching grant be greater than
  or equal to f divided by the matching grant
  multiplier (4)
• Example To take the extreme case, if f is the
  maximum of 10, and giving in the normal version
  is no less than 2.50 (10/4), giving in the
  matching grant version will be lower than in the
  normal version

44
Reconciliation of Conditional Altruism
withActual Allocation Decisions

• Modal gifts of Givers are consistent with
  hypotheses about motives in each respective
  treatment 5 in Standard, 3 in Subsidy, 5 in
  Tax 15/5, 8 in Tax 18/2 create equal splits of
  total modal transfer in Charity implies a f of
  10
• Partial crowding out in Tax and Subsidy
  treatments is consistent with Propositions 4 and
  5 (and impure altruism), but not with pure
  altruism or warm glow
• Relative magnitudes of crowding out in Tax versus
  Subsidy experiments are consistent with
  conditional altruism, but not predicted by any
  theory of unconditional altruism crowding out is
  48 in the Tax experiment and 18 (less than
  one-half) in the Subsidy experiment
• Gifts in Charity treatment are larger than in
  Standard treatment, but by less than 5
  difference in ?

45
Reconciliation with Matching Grant Results

• Choice of charity
• Increased proportion of dictators choosing
  matching grant charity is consistent with
  conditional altruism, and also pure or impure
  altruism, but not warm glow
• Level of giving
• Pattern of mean gifts in the Charity experiment
  is consistent with conditional altruism but not
  any theory of unconditional altruism
• Mean gifts in Childreach (match) version are less
  than Childreach (normal) and Children
  International (when given the choice of
  Childreach with match).

46

• Effects of matching grants on expected
  revenues of charities, i.e., donations averaged
  across the entire pool of donors
• Average Childreach donations rise from 2.57 to
  3.26 average gift per donor falls from 7.20 to
  5.43 but percentage of potential donors
  Childreach captures rises from 36 to 60
• Average Children International donations fall
  from 2.74 to 1.66 average gift per donor rises
  from 5.82 to 8.29 but capture rate falls from
  47 to 20
• Across all dictators, the matching grant reduces
  average giving from 5.31 to 4.91 (insignificant
  difference)
• Prisoners dilemma Charity offering the matching
  grant benefits while aggregate charitable
  donations do not rise and perhaps fall?

47
Reasons for Giving

• In post-experimental questionnaires, about
  two-thirds of high gifts were explained in terms
  of equity/equality in the Standard and Subsidy
  sessions and in terms of other peoples needs in
  the Charity sessions
• Low gifts were often explained by appeal to the
  dictators own needs
• Own needs were not borne out when analyzed based
  on the reported material well-being of generous
  vs. selfish dictators

48
Reconciliation of Conditional Altruism with
Results on Affective Motivation

• Conditional altruism conjecture is that better
  feelings are associated with gifts that comply
  with (beliefs about) the relevant social norm, f.
• Average dictator gifts will lie somewhere between
  impartial value and more selfish allocation.
• In ascending order, f is lowest in the Subsidy
  treatment (with an impartial f of 3), then
  Standard (impartial 5), Charity with matching
  grant (infer from average gift of 5.43 that
  impartial value is higher than 5), and Charity
  without the matching grant (10).
• Linear regression of SRAD on gifts shows
  increasing slope in same order (dark lines in
  Figure 2).
• Conclusion Better feelings are associated with
  low gifts when f is low and high gifts when f is
  high.

49

• Subsidy (a), Standard (b), matching version
  Charity (c) and normal version Charity (d)
  treatments.

50
Conjectures and Suggestive Evidence forFurther
Research on Feelings and Moral Norms

• Short run feelings are associated with social
  norm compliance as represented by the conditional
  altruism term
• Retain the term that relates to giving per se
  (like warm glow) to explain incomplete crowding
  out, but not assumed to be motivated by short run
  affect
• Long run happiness one explanation for
  incomplete crowding out is relationship of giving
  to long-run feelings giving is like an
  investment that contributes to long-run happiness
  (Konow and Earley, 2008)
• Nonlinear regression of SRAD on gift (adding
  squared gift term to right hand side) produces
  lighter, mostly concave lines

51

• Subsidy treatment (Figure 2A)
• Nonlinear regression of SRAD on gift produces
  lighter, concave line for Subsidy case.
  Self-serving bias suggests average f slightly
  less than 3.
• In other treatments, the salient values for f
  differ by a wider margin. Other studies suggest
  these beliefs cluster mostly around two values
  the impartial f and very self-serving amounts
  (0-1). Thus, the observations are partitioned
  in the other treatments (by vertical lines).

52

• Standard treatment (Figure 2B)
• Two clusters around 0 and 5 two outliers at
  10.
• Of group giving less than 3, 64 of available
  responses explained based on Own Need and none
  based on Equity.
• Of the group giving 3 to 7, 89 referred to
  fairness and none to need (no explanations
  available for two 10 gifts).

53

• Charity treatment with matching grant (Figure 2C)
• Concave SRAD in gifts greater than 1 maximized
  at 6.25.
• Anecdotal evidence on concern for efficiency 86
  of dictators who gave to Childreach with the
  matching grant pointed to the increased benefit
  as a reason for choosing that charity.

54

• Charity treatment without matching grant (Figure
  2D)
• Concave SRAD in gift greater than 1 maximized at
  14.25 (partially explains two dictators who not
  only gave their 10 endowment but also dipped
  into their show up fees?).

55
Conclusions

• This study provides evidence on altruism using a
  design that minimizes confounding effects
• Partial crowding out in the various experiments
  contradicts pure altruism and warm glow alone
• Unconditional altruism, including impure
  altruism, is also refuted by additional evidence
  on short run affect and matching grant giving
• Charity experiment indicates greater generosity
  to charities cannot be attributed primarily to
  greater familiarity
• Mixed feelings greater generosity is associated
  with worse feelings in the Standard and Subsidy
  treatments, and perhaps with better feelings only
  in the Charity treatment
• Conditional altruism offers a means to reconcile
  the otherwise contradictory results on both
  allocation decisions and affective motivation
  across the various experiments