Econometrics Project - What are the determinants of giving in Uganda

2015
Dan Drummond
Econometrics project
01/04/2015
What are the determinants of
giving in Uganda?

Dan Drummond H00124667
1
What are the determinants of giving in Uganda?
I. Introduction and motivation
This paper aim to assess the determinants of giving within the Central African nation of
Uganda. To achieve this I have used data acquired by Catherine Porter, the University of
Oxford and the Centre for the Study of African Economies (CSAE) in 2014. The data contains
sector, education, nationality, gender, work responsibilities, altruistic preferences and
charitable preferences. This is of interest as there has been very little research on dictator
games within Africa, especially with a design that caters for further giving and uses charities
as the recipient.
Giving has been well researched in the ‘West’ on student populations, however there is very
little data on African nations. The participant was given 20,000 UGX (Ugandan shillings) and
asked to divide it between themselves and a popular charity. This aimed to highlight the
characteristics of altruism in Uganda. This is a relatively large ‘windfall’ as the GNI per capita
of Uganda is $510 (World Bank, 2014). The denominations were in 1,000 UGX bills and they
had a choice of 23 charities. I have used the data to form dummy variables, I then regressed
the dummies on total given to reveal how sector employed, nationality, age or education affect
charitable donations. However total given includes more than just the 20,000 UGX as the
individuals also had the option to give their own money.
Bekkers (2007) found people who are educated are usually more prosocial and trusting. I test
this finding by looking at how their level of education changes the total donation. Sector may
also effect giving as wages and pro-social behaviour will change between sectors. Civil servants
may give less as they feel they have contributed through their underpaid work (Buurman et al,
2009).
To reach a satisfactory conclusion I assess the current literature surrounding the topic of
Dictator Games and Generosity in Section 2, this shall then be discussed further by delving
into the theory surrounding the topic in Section 3. Section 4 shall then combine these two
sections to explain my variables and give weight to the econometric model. Section 5 shall be

2
a discussion of the and the source attained, leading to Section 6 which is a discussion of the
estimation procedures, mainly the Gauss-Markov assumptions. Section 7 then details my
estimated results and their relevance. With Section 8 then pointing out the limitations of this
study and Section 9 concluding my findings. Leaving Section 10, the appendix, and Section 11,
the bibliography.
Certain characteristics determine giving within dictator games more than others. For each
year a Ugandan ages their contribution increases by 170 UGX and if they have a graduate
degree they will give 10,202 shillings more. While if they are employed in the public sector
they give 8,425 UGX less and if they are a student they give 3,654 less. While men were found
to give less, however this was insignificant. The participant’s nationality is inconclusive, with
their donation in line with economic theory but the result is insignificant at a p –value of 0.195.

3
II. Literature review
A dictator game is whereby an individual is given a ‘windfall’ and asked to divvy it
between themselves and a recipient, in this case one of 23 charities. The game is not strategy
intense since the recipient cannot veto the donation therefore the dictator’s decision is final.
Economists expect individuals to be rational and take the full endowment, this is referred to
as income maximisation, however this is rarely the case and many split their endowment.
Dictator games have been researched in great depth with over a hundred dictator games being
published (Engel, 2011). The Dictator Games and human behaviour originate from Daniel
Kahneman’s ultimatum game 30 years ago (Guth et al. 1982). This game assessed how people
behave when they are given money and have to decide on how to divide it between them. From
here stemmed the dictator game, the difference between this and the ultimatum game is that
the recipient has no say in accepting or rejecting the terms of the offer. The amount given
changes depending on certain factors, Engel found that dictator games had been adapted to
either test Situational or Demographic effects on giving. Engel’s meta study covered 131 papers
and revealed that the mean donation across 616 treatments was 28.35%. The paper also
highlighted the most significant factors across all 131 papers. He found that 63.89% violate the
income maximisation hypothesis and those who give donate an average of 42.64%.
Furthermore it highlighted that if the individual is not anonymous they feel “urged” to donate.
The majority of economic testing is done on students, an easily accessible sample, however
Carpenter et al (2007) found that students are not representative of the entire population.
Community members were found to give 17% more on average, 32% of them gave their full
endowment, and 3 times more likely to give away their full endowment. They also found that
education has a positive correlation with giving, and a positive correlation between age and
donations as well as male students being less generous.
Eckel and Grossman (1996) employed a double anonymous dictator game and tested if the
motivation for altruism increases when the recipient was replaced with a renowned charity,
American Red Cross. They found that human behaviour can be motivated by altruism and

4
further by the “deservingness” of the recipient. They concluded that “fairness and altruism
require context”.
Buurman et al (2009) tested risk averse and altruistic behaviour in public sector employees.
Their findings showed private sector workers to be more altruistic than public sector workers.
They concluded that this was because civil servants believe they’re underpaid and their job is
a substitute for charity.
Bekkers (2007) found that “generosity increased with age, education, income, trust, and
prosocial value orientation.” His research also covered current students and graduated, and
found that the effect of education on giving was only apparent post completion.
In his review article Andreoni (2006) found that higher educated individuals gave more, more
frequently and a higher fraction. He also found that age is positively correlated with generosity
and gender also effects generosity.
In their review article Bekkers and Wiepking (2007) concluded that members of the economics
faculty are more “self interested” and give less.
Tonin and Vlassopoulos (2014) found that “prosocial motivation” decreases over the length of
public sector employee’s career until it ceases to be prevalent. However they found that
occupation with in the civil service does not make effect prosocial behaviour.
Wang. M. (2013) found that the “upper socioeconomic class” were more altruistic in
comparison to the “lower socioeconomic class.” However this was from a “puzzle-liked dictator
game”, and if the individual does a task to earn their endowment it can lead to decreased
generosity. The lower socioeconomic class may feel their efforts required the reward more as
they are less income elastic.
My work shall contribute to the existing literature by applying the current dictator game and
generosity theory to a developing African nation. Checking for the significance of the theory
on the nation and then arguing for the relevance of the results or theory. I then apply my data
to the current gender and generosity debate, concluding that my data shows that gender has
an insignificant effect on giving. Followed up by recommending what could be done to further
this and current research surrounding the topic of generosity and dictator games in Africa.

5
III. Economic model
The neoclassical economic theory believes that individual’s base decisions on utility
maximisation and will therefore take the full endowment for themselves. However Engel
found that the mean split across 131 papers was 28%. This may be explained using Andreoni’s
4 reasons for altruism found in his ‘Handbook of the Economics of Giving, Altruism and
Reciprocity’:
1. Contributes to a public good they consume
2. “Enlightened self-interest”
3. Altruism due to belief in the cause
4. “Warm-glow” – people enjoy giving
The higher the individual’s disposable income the more likely they are to donate to charity.
Data for income has not been collected but income and disposable income are correlated with
age, gender, education and sector employed.
I have used the sector worked in and the individual’s education level as a proxy for wage both
have a high positive correlation with wage; average age for workers without formal education
is 546 UGX compared to university educated employees whose average earnings are 2383
UGX (WageIndicator Survey, 2012).
Civil servants earn a fair wage in Uganda with 92% earning above the poverty line
(WageIndicator Survey), compared to 24.5% of the entire country being below the poverty line
(WorldBank). Meaning they should be more altruistic than the general public as on average
they earn more. However studies have found that public servants give less as they feel their
work is contribution enough (Buurman et al).
Students have been shown to be non-altruistic, one argument for this is less life experience the
other being income (Carpenter et al,). The older the participant the more generous, this may
possibly be because of more life experience and therefore more altruistic or higher disposable
income from lifetime savings (Carpenter et al; Engel). Gender has also been shown to effect
giving, women are found to give more (Eckel and Grossman; Engel), whereas ‘Generosity and

6
Philanthropy: A Literature Review’ (Bekkers and Wiepking) had mixed findings, stating that
there was no reliable difference between genders in terms of generosity.
This literature has led me to the following hypotheses’:
Null hypothesis: H0 Alternative: H1
Age Age does not affect total given Change in age causes a change
in total given
Gender Females do not give more than
men
Females give differently to
men
Education A post graduate degree does
not affect total given
A post graduate degree
changes total given
Sector Civil servants will not have
different giving habits
Civil servants giving will be
different
Students Students giving will not differ
from that of the general public
Students will give differently to
the general public
Nationality Ugandans won’t give
differently in comparison to
other nationalities
Ugandans will give differently
in comparison to other
nationalities

7
IV. Econometric model
𝑡𝑜𝑡𝑔𝑖𝑣𝑒 = 𝛽0 + 𝛽1 𝐴𝑔𝑒 + 𝛽2 𝑀𝑎𝑙𝑒 + 𝛽3 𝑃𝑜𝑠𝑡𝐵 + 𝛽4 𝐶𝑖𝑣 𝑆𝑒𝑟𝑣 + 𝛽5 𝑆𝑡𝑢𝑑𝑒𝑛𝑡
+ 𝛽6 𝑈𝑔𝑎𝑛𝑑𝑎𝑛 + 𝑢
Age
Age is the age of the questionnaire respondent in years. The range for age was 20 to 63 with
only 2 individuals not answering. The mean age was 32.14 and the median age was 30. 77/148
individuals were in the age group 20-30. This variable is included as both Engel and Carpenter
et al found that age was significant in altruism.
Male
Male is the dummy variable for the gender of the individual, with 1 being male and 0 being
female. With 80/148 participants are male therefore making the mean 0.544, and baseline is
female. This variable is included as there’s contradictory findings surrounding gender, with
Eckel and Grossman and Engel finding women give more and Bekkers and Wiepking finding
gender insignificant.
Education
Symbol Variable
Post_B Education beyond a Bachelor’s degree (e.g
Masters or PhD)
This dummy variable has been created from the questionnaire, the question stated 9 different
levels of education from 1 being none to 9 being a PhD, I used this to create 3 Dummy variables.
High School or lower, with only one individual not attending high school, 30 finishing high
school, and the rest were higher education. Bachelor’s consisting of 88 individuals who had
done a Bachelor’s degree. Post_B was 29 participants who had a graduate degree, everything
below Post_B is the baseline. Bekkers, Bekkers and Wiepking and Andereoni all find a positive
relationship between education and philanthropy.
Sector

8
Symbol Variable
Civ_Serv Civil Servant
Student Academic Student
The two sectors are dummy variables attained from the results of the questionnaire. The
results are from 1 to 10, with 1 being private firm, 2 to 4 being a form of Civil Servant, 5 an
NGO worker, 6 Self-employed, 7 Volunteer, 8 not employed, 9 Other and 10 being a Student.
I then obtained 4 dummies consisting of 2 sets of group data. I did not use the NGO and
Volunteers, nor 1, 6, 8 or 9, thus making them my base line. 55/148 of the respondents were
Civil Servants and 40/148 was students. Students and civil servants have been included as
they’ve been found to be less altruistic (Carpenter et al; Buurman et al).
Ugandan
This is a dummy variable based on the individual’s nationality. If the individual is Ugandan it
is 1, if not it is 0. The mean was 0.716, the other nationalities are the bench mark. This is
included in the model as Engel’s Meta Study found that non-western nations gave more.
I have avoided the dummy trap as I have baselines for all my dummies. The model is also a
level/level model and all figures are in UGX.

9
V. Data
This data was collected by Catherine Porter, the experiment measured how much of
the 20,000 UGX endowment the 148 recipients gave to charity. The data was collected from 3
separate locations, an up market coffee shop, Ministry of Finance and Economics students at
the Makerere University. It consisted of a dictator game whereby the subject was given two
envelopes, one that read “take with you” and one that read “leave with research team”. The
respondent received 20,000 UGX (approximately £4.51 [XE.com, 2015]). The individual filled
out a questionnaire consisting of 12 questions reflecting demographic, social, economic and
behavioural characteristics. Then the individual chose if they wanted to donate to charity and
if so which of the 23 charities they wanted to donate to. Of the 23 charities, 8 were Ugandan,
11 were children orientated, 15 were international charities and 2 were religious. The data was
inserted into excel, where I created the necessary dummy variables, then exported into EViews
as to enable regression.
The average donated was 9845 UGX with a minimum of 0 and a maximum of 40,000. Only
16% (23/148) of the subjects were pure gamesman (did not give any of their
endowment/income maximisers). 7 individuals gave above 20,000 with 1 giving 30,000 and
4 giving 40,000. This pulled the mean substantially above the median, 8,000.
The distribution (Figure 1) is partially in line with Engel’s description of a two peaked
distribution, with one peak representing pure gamesman and the other the equalitarian
option. The first peak for ‘give nothing’ can be seen, however the second peak is at the full
endowment with a very small peak at the equal split. The difference may be due to our
participants having the option to give more than the full endowment. The distribution does
show that the data is slightly positively skewed at 1.04, skewness is a measure of asymmetry
of the variable’s distribution around its mean.
The standard deviation measures the asymmetry of data around the mean. The standard
deviation is 9356; showing that the data is highly asymmetrical due to the large spread.

10
Non Binary:
Variable Definition Mean Minimum Maximum Std. Dev
Totgive Total Given 9,844.595 0 40,000 9,355.934
Age Age in years 32.143 20 63 10.119
Binary Variables:
Variable Definition Mean Fraction Std. Dev
Male Of male gender 0.544 80/147 0.5
Post_B Has a Graduate degree 0.196 29/148 0.4
Civ_serv Civil Servant 0.372 55/148 0.485
Student Currently a student 0.27 40/148 0.446
Ugandan Of Ugandan nationality 0.716 106/148 0.452

11
VI. Estimation procedures
The model is a level-level model and was estimated using OLS in EViews. OLS is an
appropriate method of estimation as it minimises the difference between the responses and
the predicted responses by linear approximation.
For the specification of my model I chose to follow economic tuition (based on my literature
review) over statistical significance (Ramsey RESET test). As I believe statistical significance
and specification can lead to data mining.
The Best Linear Unbiased Estimators (BLUE) has the following 5 conditions:
1. Linear in parameters
2. Random sampling
3. No Perfect collinearity
4. Zero conditional mean
5. Homoskedasticity
Linearity: 𝑦 = 𝛽0 + 𝛽1 𝑥 + 𝑢
The model is linear as it is a linear combination of observed values for the dependent variable
(totgive).
Random Sampling: 𝑛{(𝑥𝑖1 … 𝑥𝑖𝑘, 𝑦𝑖): 𝑖 = 1,2 … 𝑛}
The population is random in sample because individuals were picked randomly, however the
locations were not. Also the variation in variables does not equal zero and are therefore
random.
No Perfect Collinearity: ∑ (𝑋𝑖 − 𝑋̅𝑛
𝑖=1 )2
≠ 0
The Spearman Rank (Figure 2) has one value above 0.5, with student and age having a
collinearity of -0.618. Therefore I do not have perfect collinearity as I can regress the equation.
However I have high collinearity, this is expected as 35 of the 54 individual aged 20-25 are
students (64.8%).

12
Zero Conditional Mean: 𝐸(𝑢𝑖|𝑥𝑖) = 0
𝐻0: 𝜎1 = 𝜎2 = 0
𝐻1: 𝜎1 ≠ 𝜎2 ≠ 0
If 𝐻 𝑜 is correct I have misspecification
If 𝐻1 reject 𝐻 𝑜
Ramsey Reset Test
F-statistic 2.43506 Prob. F(2,136) 0.0914
Log likelihood ratio 5.101599 Prob. Chi-Square(2) 0.078
Misspecification is when you fail to include a relevant independent variable or use an incorrect
functional form. I fail to reject the null hypothesis at the 0.1 p-value (10% significance),
therefore my model is misspecified. The Ramsey RESET tests for statistical specification
whereas I am specifying according to economic intuition. Therefore I am going to assume that
my model is correctly specified as I do not have theory to support adding any additional
variables that I possess. My model is also in correct form or would not regress.
MLR1-4 hold therefore we have unbiased estimators.
Heteroskedasticity: 𝑉𝑎𝑟(𝜀|𝑋𝑖) = 𝜎𝜀
2
∀𝑖
𝐻0 ∶ 𝛾1 = 𝛾2 = … = 𝛾𝑘 = 0
𝐻1 ∶ 𝐻0 𝑖𝑠 𝑖𝑛𝑐𝑜𝑟𝑟𝑒𝑐𝑡
If 𝐻 𝑜 is correct you have homoscedasticity
If 𝐻1 reject 𝐻 𝑜
White test (Figure 3): F Stat: 2.43 and p value: 0.0014
BP Test (Figure 4): F Stat: 3.14 and p value: 0.0065
Therefore both tests show heteroskedasticity, as I fail to reject the null hypothesis at a p-value
of 0.1. So I’ve created a Heteroskedasticity robust output (Figure 5), therefore MLR.5 holds
and thus my estimators are BLUE.

13
VII. Estimated results
Heteroskedasticity robust results:
Variable Coeffecient t-Stat
Age 170.42** 2.110081
Male -672.09 -0.55289
Post_B 10201.71* 4.116463
Civ_Serv -8425.004* -3.52358
Student -3654.291*** -1.74938
Ugandan 2971.743 1.3037
*Significant at 1%
**Significant at 5%
***Significant at 10%
Ugandan and Male have been kept in because of economic theory. All other variables
are significant within the recommended 10% level.
Age has a positive effect, if all other variables are held constant (ceteris paribus) then for each
additional year the individual donates 170.42 UGX more, significant at the 5% level. Engel
found that the middle aged very rarely give nothing and the elderly never give nothing. The
life expectancy in Uganda in 2012 was 58.7 (Unicef, 2012), therefore I define Middle aged as
40 and elderly as 55 and older. None of the individuals older than 37 gave nothing, and only
one individual above 35 gave nothing (35 and above accounts for 44/146). This is in line with
economic theory. Figure 7 displays a scatter of Age and Totgive; the trend line shows the
progressive increase of donations across age.
Male has negative effect as under ceteris paribus if the individual is male they give 672.09 UGX
less, insignificant at the 10% level. Engel found that women were more generous, our data
displays a similar result. However this variable is not significant, and is therefore in line with
Bekkers and Wiepking’s findings.
Post_B has a positive effect as under ceteris paribus if the individual has graduate degree they
give 10,201.71 UGX more, significant at the 1% level. Bekkers found that graduates had
increased generosity in comparison to those of lower educational status, theorising that this
be due to greater income, prosocial value orientations and trust. This was supported by

14
Andereoni who found that higher levels of education gave more, more frequently and a greater
fraction.
Civ_Serv has a negative effect as under ceteris paribus if the individual is a Civil Servant they
give 8425.004 UGX less, significant at the 1% level. Buurman et al found that civil servants are
less inclined to donate than private sector employees, my results support this with civil
servants give less than the normal citizens, and students.
Student has a negative effect as under ceteris paribus if the individual is a student they give
3654.291 UGX less, significant at the 10% level. Carpenter et al found that students are 32%
less likely to give the full endowment. Out of those who gave everything 6/44 were students,
lower than Carpenter’s. Engel found similar; students more likely to give nothing, less likely
half the endowment or to give the full endowement. Of those who were pure gamesman and
gave nothing 9/23 were students, and 4 were civil servants.
Ugandan has a positive effect as under ceteris paribus if the individual is Ugandan they give
2971.743 UGX more, insignificant at the 10% level but has economic theory. Engel found a
correlation between the level of development and altruistic behaviour, with indigenous
communities being the most generous. Kampala is the capital, and cannot be categorised as
indigenous but it can be categorised as developing.

15
VIII. Limitations of the study
Khaneman and Tversky (1971) proposed a law of small numbers, this law required
there to be 150 or more subjects for the sample to be representative of the population. Our
sample only has 148, therefore the law does not hold and this cannot be seen as representative
sample.
The sample is of higher income individuals and not representative of all the population. The
sample was held in three locations leading to some level of sample bias. The locations were an
Economics department in a university, the ministry of finance and a high end coffee shop.
Economics students and professors have been found to be self-interested.
A small sample means that the standard deviation is larger leading to inaccurate results. As
the standard deviation is the difference each sample is away from the mean, since the lower
the sample the larger the difference between means. This leads to a wider confidence interval
and hence significance level of 10%.
Kampala is the capital of Uganda, therefore the average demographics are higher than the rest
of the country. The sample is therefore not representative of the population of Kampala as the
sample has higher income, more education and better employment. The sample is also not
representative of Uganda as it is within Kampala where the demographics are higher than that
of the rest of the country.
The use of a questionnaire creates measurement error through non response bias and bias
from the design of the questionnaire and questions, such as the wording or weighting of the
questions. This could of lead to possible measurement error, such as the confusion with
question 12, whereby people were asked if they would like to give their own money whereby
36 said yes but only 9 actually gave their own money. If the data input into the model is bias
then the results of the model will also be bias. The questionnaire had errors, with 2.7% of
people not turning the page over and 10.8% of people not answering question 8. The

16
questionnaire may also induce bias as the envelope may decrease the amount of pure
gamesman as lack of anonymity changes their donations (Eckel and Grossman).
The model would be more accurate with the addition of income. With Q7 being a potential
proxy for income, it asks for the level of responsibility within the individuals role. This is not
a great proxy for income as different sectors will pay different amounts and students will not
answer the question, however they still have an income. Also the pay scale for civil servants is
publicly available and could have been featured in the questionnaire as 37% of the sample are
civil servants. I have included sector within my model, this will account for some of the
variation in income as the private sector wage will differ from that of the public, however I
have not used it as a proxy for income as it is too ambiguous.

17
IX. Conclusion
Variable Null
hypothesis:
H0
Alternative:
H1
Coefficient t-Stat Result
to H0
Interpreta
tion
Age Age does not
affect total
given
Change in
age causes a
change in
total given
170.42** 2.110 Reject Age
positively
affects the
amount
given
Gender
(Male)
Females do
not give more
than men
Females
give
differently
to men
-672.09 -0.553 Fail to
reject
Gender
does not
affect total
given
Education
(Post_B)
A post
graduate
degree does
not affect
total given
A post
graduate
degree
changes
total given
10201.71* 4.116 Reject Having a
graduate
degree has
a strong
positive
effects on
giving
Sector
(Civ_Serv)
Civil servants
will not have
different
giving habits
Civil
servants
giving will
be different
-8425.004* -3.524 Reject Being a
civil
servant
has strong
negative
effects on
giving
Students
(Student)
Students
giving will
not differ
from that of
the general
public
Students
will give
differently
to the
general
public
-3654.291*** -1.749 Reject Being a
student
has a
negative
effect on
giving
Nationality
(Ugandan)
Ugandans
won’t give
differently in
comparison
to other
nationalities
Ugandans
will give
differently
in
comparison
to other
nationalities
2971.743 1.304 Fail to
reject
Being
Ugandan
does not
affect
giving
*Significant at 1%
**Significant at 5%
***Significant at 10%
I have found that Age, graduate degree, sector employed and whether or not a student
effect the amount individuals in Kampala, Uganda give by varying amounts. As age increases
the participant gives significantly more and if the individual has a graduate degree they give
substantially more. While if they are employed in the public sector they give 8,425 UGX less

18
and if they are a student they give 3,654 less. I thought that civil servants would give less than
the rest of the public but more than students. This may be because civil servants believe their
job gives enough to society.
Gender was insignificant, there was contradictory findings for this and my findings support
Bekkers and Wiepking’s findings. Nationality was insignificant but the findings were in line
with Engel’s theory of developing nations and donations. The sample was not representative
of Uganda or Kampala, this limits the further use of the study.
Further research should aim to be more representative of geographic regions or certain
demographics and include income. This may allow for a significant result for nationality which
may support or contradict Engel’s theory. Income would also give us a more representative
data on altruism in Africa, as we would have amount given relative to income.

19
X. References
Andreoni, J. (2006) ‘Philanthropy’, in Kolm, S.-C. and Ythier, jean M. (eds) Handbook of
the Economics of Giving, Altruismand Reciprocity, Volume 2.
Bekkers, R. (2007) ‘Measuring Altruistic Behavior in Surveys: The All-or-Nothing Dictator
Game’, Survey Research Methods, 1.
Bekkers, R. and Wiepking, P. (2007) ‘Generosity and Philanthropy: A Literature Review’,
Science of Generosity. doi: 10.2139/ssrn.1015507.
Besamusca, J. and Tijdens, K. (2012) Wages in Uganda: WageIndicator survey 2012.
Buurman, M., Dur, R. and van den Bossche, S. (2009) ‘Public Sector Employees: Risk Averse
and Altruistic?’, SSRN Electronic Journal. doi: 10.2139/ssrn.1441844.
Carpenter, J., Connolly, C. and Myers, C. K. (2007) ‘Altruistic behavior in a representative
dictator experiment’, Experimental Economics. Springer, 11(3), pp. 282–298. doi:
10.1007/s10683-007-9193-x.
Eckel, C. C. and Grossman, P. J. (1996) ‘Altruism in Anonymous Dictator Games’, Games
and Economic Behavior, 16(2), pp. 181–191. doi: 10.1006/game.1996.0081.
Engel, C. (2011) ‘Dictator games: a meta study’, Experimental Economics. Springer, 14(4),
pp. 583–610. doi: 10.1007/s10683-011-9283-7.
Guth, W., Schmittberger, R. and Schwarze, B. (1982) ‘An experimental analysis of ultimatum
bargaining.’, Journal of Economic Behaviour and Organization, 3.
List, J. A. (2007) ‘On the Interpretation of Giving in Dictator Games’, Journal of Political
Economy, 115(3), pp. 482–493. doi: 10.1086/519249.
The World’s Trusted Currency Authority (no date). Available at: xe.com (Accessed: 31 March
2015).
Tonin, M. and Vlassopoulos, M. (2014) ‘Are Public Sector Workers Different? Cross-
European Evidence from Elderly Workers and Retirees’, Institute for the Study of Labour,
8238.
Tversky, A. and Kahneman, D. (1971) ‘Belief in the law of small numbers.’, Psychological
Bulletin, 76(2), pp. 105–110. doi: 10.1037/h0031322.

20
Uganda (no date). Available at: http://data.worldbank.org/country/uganda (Accessed: 5
March 2015).
‘Uganda Statistics’ (2013). UNICEF. Available at:
http://www.unicef.org/infobycountry/uganda_statistics.html (Accessed: 10 March 2015).
Wang, M. (2013) ‘Does higher socioeconomic class predict increased altruistic behavior?
Evidence from a modified dictator game’, Erasmus University Rotterdam.

21
XI. Appendix
Figure 1: Dot plot of totgive data
Figure 2: The distribution of the dependent variable totgive (total given)
0
10,000
20,000
30,000
40,000
50,000
25 50 75 100 125
TOTGIVE
0
10
20
30
40
50
60
0 10000 20000 30000 40000
Series: TOTGIVE
Sample 1 148
Observations 148
Mean 9844.595
Median 8000.000
Maximum 40000.00
Minimum 0.000000
Std. Dev. 9355.934
Skewness 1.037433
Kurtosis 3.823861
Jarque-Bera 30.73352
Probability 0.000000

22
Figure 3: Spearman Rho rank displaying the correlation of my independent variables
Heteroskedasticity Test: White
Obs*R-squared 42.49408 Prob. Chi-Square(21) 0.0036
Scaled explained SS 85.68952 Prob. Chi-Square(21) 0.0000
Test Equation:
Dependent Variable: RESID^2
Method: Least Squares
Date: 03/17/15 Time: 20:32
Sample: 1 148
Included observations: 145
Collinear test regressors dropped from specification
Variable Coefficient Std. Error t-Statistic Prob.
C -8192379. 1.57E+08 -0.052055 0.9586
AGE 2303469. 8320087. 0.276856 0.7824
AGE^2 16331.11 115068.9 0.141925 0.8874
AGE*MALE -120890.1 2685683. -0.045013 0.9642
AGE*POST_B -9873001. 3362552. -2.936164 0.0040
AGE*CIV_SERV -3890594. 3111609. -1.250348 0.2135
AGE*STUDENT 9490386. 9760562. 0.972320 0.3328
AGE*UGANDAN -706674.3 3255570. -0.217066 0.8285
MALE -29794966 95988055 -0.310403 0.7568
MALE*POST_B 1.18E+08 57259238 2.053714 0.0421
MALE*CIV_SERV -81058268 73220551 -1.107043 0.2704
MALE*STUDENT -95456320 60642014 -1.574095 0.1180
MALE*UGANDAN 1.20E+08 66247017 1.817495 0.0716
POST_B 2.47E+08 1.28E+08 1.934782 0.0553
POST_B*CIV_SERV 1.29E+08 82709973 1.564090 0.1204
POST_B*STUDENT -1.29E+08 1.38E+08 -0.930111 0.3541
POST_B*UGANDAN 63488120 76884447 0.825760 0.4105
CIV_SERV 15452240 1.63E+08 0.094613 0.9248
CIV_SERV*UGANDAN 57310660 1.43E+08 0.399641 0.6901
STUDENT -1.38E+08 2.40E+08 -0.574752 0.5665
STUDENT*UGANDAN -76357162 67287360 -1.134792 0.2587
UGANDAN 9129231. 1.11E+08 0.082125 0.9347
R-squared 0.293063 Mean dependent var 55129806
Adjusted R-squared 0.172366 S.D. dependent var 1.17E+08
S.E. of regression 1.06E+08 Akaike info criterion 39.93839
Sum squared resid 1.39E+18 Schwarz criterion 40.39003
Log likelihood -2873.533 Hannan-Quinn criter. 40.12191
F-statistic 2.428093 Durbin-Watson stat 2.220192
Prob(F-statistic) 0.001354

23
Figure 4: White test for heteroskedasticity within my model
Heteroskedasticity Test: Breusch-Pagan-Godfrey
Obs*R-squared 17.39438 Prob. Chi-Square(6) 0.0079
Scaled explained SS 35.07585 Prob. Chi-Square(6) 0.0000
Test Equation:
Dependent Variable: RESID^2
Date: 03/05/15 Time: 18:56
Sample: 1 148
C 52122176 43621273 1.194880 0.2342
AGE -605395.0 1135303. -0.533245 0.5947
MALE 25736757 19779629 1.301175 0.1954
POST_B 41373327 26357811 1.569680 0.1188
CIV_SERV -84348153 27517607 -3.065243 0.0026
STUDENT -58758945 27772251 -2.115743 0.0362
UGANDAN 67013658 25890775 2.588322 0.0107
R-squared 0.119961 Mean dependent var 55129806
Adjusted R-squared 0.081699 S.D. dependent var 1.17E+08
S.E. of regression 1.12E+08 Akaike info criterion 39.95051
Figure 5: Breusch-Pagan test for heteroskedaticity within my model
Dependent Variable: TOTGIVE
Date: 03/17/15 Time: 21:07
Sample: 1 148
White Heteroskedasticity-Consistent Standard Errors & Covariance
AGE 170.4154 80.76252 2.110081 0.0367
MALE -672.0931 1215.606 -0.552887 0.5812
POST_B 10201.71 2478.271 4.116463 0.0001
CIV_SERV -8425.004 2391.034 -3.523582 0.0006
STUDENT -3654.291 2088.909 -1.749378 0.0824
UGANDAN 2971.743 2279.468 1.303700 0.1945
C 4642.149 2798.224 1.658963 0.0994
R-squared 0.367037 Mean dependent var 9800.000
Adjusted R-squared 0.339516 S.D. dependent var 9364.976
S.E. of regression 7610.929 Akaike info criterion 20.75963

24
Figure 6: White heteroskedasticity robust output for my model.
Figure 7: Scatterplot for age and total given with trendline
0
5000
10000
15000
20000
25000
30000
35000
40000
45000
0 10 20 30 40 50 60 70
totgive

Econometrics Project - What are the determinants of giving in Uganda

More Related Content

What's hot (10)

Viewers also liked (9)

Similar to Econometrics Project - What are the determinants of giving in Uganda (17)

Econometrics Project - What are the determinants of giving in Uganda