Attitudes towards income inequality: A case-study performed in Voi division, Kenya

1

UPPSALA UNIVERSITY Department of Economics Bachelor thesis

Author: Maria Carlsson Supervisor: Jan Pettersson Autumn term, 2011

Attitudes towards income inequality

- A case-study performed in Voi division, Kenya

2

Abstract

The actual outcome in terms of inequality as well as the expected effect from an income distribution depends on the view in the society about how to allocate the income. This essay tries to investigate how gender, current income, trust and Hirschman‟s tunnel effect theory can explain people‟s attitudes towards income inequality. The data used in this essay was collected in Voi division, Kenya, a place with a relatively high rate of income inequality and a place where attitudes towards income inequality has not been investigated before. The data collected was analyzed with ordinal logistic regression and the result shows that gender, satisfaction with current situation, experience of control and free choice regarding the future, marital status, if the respondents have children, if the respondent believe in god and attitudes towards ethnical diversification is variables that significantly impact on people‟s attitudes towards income inequality.

Keywords: attitudes towards income inequality, ordinal logistic regression, gender, current

income, future mobility, trust

3

Index

1. Introduction ... 4

2. Conceptual framework ... 7

3. Data procedure ... 8

3.1 Collection of data ... 8

3.2 Variables ... 11

4. Econometric model ... 15

4.1 Ordinal logistic regression ... 15

4.2 Interaction effects ... 16

5. Result ... 17

6. The state as a redistributional institution ... 21

7. Conclusions ... 24

7. References ... 27

Appendix 1 ... 29

Appendix 2 ... 35

4

1. Introduction

Economic inequality considers variations in living standards across a population and it measures how economic welfare generated in an economy is distributed among its

inhabitants. Inequality is a significant factor in developing processes and matters for, among others, economic growth, where there often is a measured correlation between future

economic growth and a smaller share of income inequality. It is also supposed to constitute a factor in economic- and social development as well as for poverty-reduction. (Perkins et al., 2006)

But even if a smaller amount of income inequality is supposed to have a positive impact on development, income inequality is present in several developing countries and cross-country comprehensions made with the Income Gini coefficient shows that developing Sub Saharan African countries, together with parts of Latin America, holds almost all top notations when it comes to measured inequality in income distribution (UNDP, 2010).

The state and the market are viewed as the primary institutions to distribute income in a modern society. The decision about how to distribute the income depends on different factors such as political incentives, governmental policies, legislation and the size of the income to distribute. But the actual outcome in terms of inequality, as well as the expected effects from an income distribution, does also depend on the view in the society about how income should be allocated. Even if income is distributed similarly in two societies, the consequences of inequality are affected by how the inhabitants in the society experience income inequality. (Szirmai, 1988)

One common conception is that attitudes towards income inequality primarily are determined by current income level, where redistribution from the rich to the poor will be supported by the rich and disapproved by the poor. But this is not necessarily true; other factors such as information shortage of peoples own position in the income ladder, as well as their future mobility expectations, is supposed to matter for how people develop their attitudes towards income inequality.

In Kenya, the country this essay will investigate, income inequality is prevalent with a

measured Income Gini coefficient on 47.7 in 2009 (this is comparable whit e.g. Sweden,

which had a measured Income Gini coefficient on 25.0 in 2009) (UNDP, 2009). The unequal

income distribution can, to a large extent, be explained by factors such as gender, in which

sector people are employed and were in the country they are settled. A typical poor person in

Kenya is a woman, employed in the agricultural sector and settled in the northern part of the

5

country. Simultaneous, a typical well-of person is a man, who works in manufacturing and settled in the central province (SID, 2006). When discussing inequality, following the concept in Amartaya Sen‟s (1985) capability approach, a distinction between inequality in

opportunities and inequality in outcomes is made. That is, factors and processes called opportunities, such as gender, geographical position and which ethnic group the person

belongs to, impact on the outcome in terms of inequality. The outcome, in turn, is what people actually get access to and how this access differs between individuals and groups. It can be outcomes in education as well as in health care or income.

One difference in opportunities is related to gender. Gender inequality is prevalent in several different sectors in Kenya. Tradition and prejudice as well as social, economic and political interests co-operate to relegate women‟s status. Women do experience a more daunting situation compared to men. The male in a household is more often seen as the breadwinner as well as the head of the household and women are not as likely as men to have a non-agricultural wage employment. In general, women get less education and furthermore, it is easier for a man than for a woman to be employed in the formal sector. Property rights are often defined in advantages of men and the majority of the female labor goes unrecorded in the system of national accounts. The government of Kenya has made some effort to promote women‟s rights and e.g. the constitution makes no difference between men and women when it comes to protecting human rights. But still discrimination is widespread and prevalent. (Marinda, 2006: Ellis et al., 2007)

The data used in this essay were collected in Voi division, which belongs to Taita district, a part of Kenya were the poverty level is 1.26 times higher than the national poverty level. In Taita district, approximately 76% of the population lives in rural areas and are dependent upon agriculture. In 2007, 58% of the population in Voi division lived below the poverty line and the poverty was more significant in the rural areas than in Voi town.

(Republic of Kenya, 2009)

The aim of this essay is to investigate how attitudes towards income inequality, one

factor used to explain the outcome in terms of income inequality, can be expounded in Voi

division. Recently, interest has been set to the prevalence and the effects of inequality in

Kenya, but underlying causes and factors regarding income inequality have not been

discussed quite as much. The main purpose with this essay is to see how attitudes towards

income inequality in the division can be explained by gender and how gender, in turn, impact

on commonly used economic explanation models for attitudes towards income inequality. The

take-off point will be in current economic research, with attitudes towards income inequality

6

related to risk taking and trust, and in Alfred Hirschman‟s tunnel effect theory. Since it appeared to be difficult to incorporate a relevant question regarding risk aversion in the questionnaire, I decided to only investigate the relationship between trust and attitudes towards income inequality. The link between trust and inequality aversion is not as well- established as the link between risk aversion and inequality aversion respectively risk aversion and trust, but since there are many studies which shows a correlation between income inequality and trust, the relation is although relevant to investigate. I also want to see if there are any indications in the data for the inherent assumption that those who are currently better-off are more likely to support income differences than those who are currently less well-off. Therefore I will add income per person as a dependent variable in the questionnaire.

I also intend to test if there are any interaction effects between gender and trust as well as between gender and future expectations and I want to see how these interaction variables impact on attitudes towards income inequality.

My expectation is that female respondents are more inequality averse than their male counterparts, since the female group is supposed to suffer more from inequality in the starting position and furthermore are less supposed to experience future mobility. I also expect both the trust variable and the variable control and experience of free choice of how the life turns out to impact positively on attitudes towards income inequality. Whether the satisfaction index is positively or negatively correlated with the dependent variable is more unclear and this variable should mainly be seen as a complement to the control and free choice variable.

This is because I want to relate the respondent‟s prospects to their current situation by making an interaction variable between the satisfaction index and the control and free choice variable.

The paper is structured as follows: First I introduce the theories used in the essay. This part follows by an introduction to the data used in the thesis. In the data section is the

variables used in the regression model also introduced. Thereby follows an introduction to the

econometric methods and subsequently is the result presented and interpreted. The result

consists of four different regression models, one model without any interaction variables

added and three regression models with different interaction variables added. The result part

follows by a robustness test where I try to see if the theoretically asked question about how

income should be distributed is likely to have any practical implementations. After that

conclusions are drawn based on the gained results.

7

2. Conceptual framework

In economic theory risk aversion and inequality aversion are commonly analyzed with the same theoretical approach. When people are uncertain about their utility function as well as their economic situation in relation to other people, due to the assumption that people in general are risk avert, the individual is assumed to choose a consumption point which is independent of which income level the person belong to. That is, a risk averse person want to have the same consumption irrespective of if it belongs to a higher or lower income level. (see e.g. Amiel et al. 1999b; Ben-Ner & Putterman, 2001; Cowell & Cruces, 2004; Davidovitz &

Kroll 2004; Morgan, Katz & Rosen, 2009). It is also supposed to be a relation between risk aversion and trust, where the correlation often is implemented as a decision by an individual whether to trust another person or not by placing a bet on the trustworthiness of an

anonymous counterpart in a situation where both parts can gain from the situation. (Ben-Ner

& Putterman, 2001; Eckel & Wilson, 2004) The link between trust and inequality aversion is, as said, not as theoretically well-established as the link between risk aversion and inequality aversion respectively risk aversion and trust, but in empirical research a link between income inequality and trust has often been established. (see e.g. Knack & Keefer, 1997; Zak &

Knack, 2001; Bjørnskov, 2007)

Alfred Hirschman‟s tunnel effect theory is another explanation model for how people develop their attitudes towards income inequality. The theory shows that not only today‟s income matters, also expected income in the future and past mobility experience impact on people‟s attitudes towards income inequality. Hirschman‟s theory has been used to explain why people, even those who suffer from rising inequality, might accept increased inequality for a while. To illustrate the theory Hirschman used a picture of a traffic jam in a tunnel:

Initially there is traffic-jam-caused stop in a tunnel and no cars moves in either direction.

People do not know how long the tunnel is and for how long the jam will persist. As a

consequence, they will start to feel resigned after a while. But, when the jam relieves and the

cars in one queue start to move, people in the other queues will immediately have a more

positive view of their own prospects and think that they will soon be out of the tunnel. This

picture can be transferred to show how people prospect their future income situation. That is,

if people expect themselves to gain from a move towards a more unequal income distribution,

tolerance for rising inequality can be present even among people who do not gain from it

initially. (Hirschman,1973)

8

3. Data procedure

This essay consists on data collected from 180 respondents in Voi Division, Taita district, Kenya. The survey was held amongst a representative sample of adult inhabitants in the district and the respondents have answered a questionnaire consisting of 27 questions. The dependent variable is measured attitudes towards income inequality and the independent variables are gender, trust, the respondent‟s satisfaction with its current life and economic situation, experience of control and free choice over its future life and income per person in the household. Different control variables have also been added to the regression.

3.1 Collection of data

The data used in this essay was collected in three different places in Voi division during May and June 2011. 90 questionnaires were answered by people living inside Voi town, 20 of the questionnaires were handed out to households near Voi River, among the Mombasa-

Taveta/Tanzania-highway, and 70 questionnaires were handed out to people in three villages

close to Maungu, a small town among the Mombasa-Nairobi-highway (figure 2). The names

of these three villages were Marungu, Kale and Mkamenyi. Since there are no strict borders

between these three villages, it is unnecessary to specify how many questionnaires I handed

out in each village.

9

Figure 1. The places where the questionnaires were handed out are marked on the map. Voi is the big, dot (indicated with an arrow) and the rural areas are the surrounding, smaller spots

.

In the collection of data I had an ideal to collect a foundation of data which was as

representative as possible for Voi division, but several times the reality forced me to accept

retrenchments in this ideal. Therefore there might be a selection bias as an effect of the

difficulties related to the collection of data. The main problem was to find a working sample

frame to draw a simple random sample from. The best would have been to choose people

from a representative index (Dahmström, 2005), but since it was impossible for me to get

access to any indexes, I tried to get a representative sample as possible by visiting different

work places, meeting places as well as randomly chosen houses within the town and in the

villages instead.

10

All participation in the study was voluntary and three people refused to answer the questionnaire. The complete non-response consisted of five people, the three who refused to answer the questionnaire and two who did not return the questionnaire to me. There is also a partial nonresponse where people who agreed to answer the questionnaire refused to answer one questions or more of the, in the regression, included variables. When a respondent did not answer one question or more this individual has been left out of the regression. Altogether, 18 individuals who answered the questionnaire were excluded from the regression model

depending of a partial non-response for one question or more. The complete non-response level is on 10.89 %. Before I started to use the questionnaire a pre-study was made with ten respondents who answered the draft. These drafts were not included in the final data set, since some questions were modified after the pilot study.

The language constituted a problem, while some people preferred to fill in the questionnaire in English and some preferred to have it in Swahili. I had a translation of the questionnaire made by two persons who are fluent in Swahili. One of them translated it for me and the other person discussed the meaning of it with me and translated it back to English again. This was made to make sure the validity and reliability in the translated questionnaire, as well as the equivalence between the original questionnaire and the translated one. Both translators were aware of my intentions with the study and the questions asked in the

questionnaire. Before I started to use the translated questionnaire a Kenyan friend pre-tested it and discussed it with me (Harkness, 2003). Because of illiteracy some questionnaires had to be read up for the respondents and I also had an interpreter with me if the questionnaire had to be read up in Swahili or if those respondents who did not speak any English had questions related to the study.

The decision to use questionnaires instead of making interviews was chosen for practical reasons. In some work places people could not answer my questions during their work day and since it was too dangerous for me to be outside in the evenings I used

questionnaires to make sure that even these persons could participate in the study. However, most of the time I could be present when the respondents answered the questionnaires and in those cases when the respondents had to fill in the questionnaires alone I was available on the phone if they had any questions about the questionnaires or the study. Altogether there were 17 persons who filled in their questionnaires without me being present.

The questionnaire (appendix 1) used in the study consists of 27 questions and is a

structured questionnaire based on the questionnaire World Values Survey (WVS) uses in their

data collection. WVS is the largest survey project to analyze values and cultural changes

11

worldwide. Their data has been collected in four different, so-called, waves and contains data from 53 different countries with different level of development and welfare. WVS have three types of questionnaires used for their data collection and I took my questions from the non OECD 2005-questionnaire (World Values Survey, 2011). The reason why I collected my own data instead of relying on WVS‟s data set is that WVS has not collected any data from Kenya yet. Neither could I find any other database Kenya with collected data regarding my purposes.

3.2 Variables

The question which is used to investigate how people value income inequality is formulated in the following way:

What do you think about the following issue? Place it on the scale below.

We need larger income differences as Income should be made more equal incentives for individual effort

1 2 3 4 5 6 7 8 9 10

If the way above is the best way to formulate the issue about income distribution is not necessarily sure. A question can be formulated in several different ways and the way the question is formulated is likely to impact on how the respondent views the issue. Therefore it is important to be aware of different nuances in the formulated question (Dahmström, 2005).

Here, a condition is added to the inequality friendly statement, which might impact upon the respondents answer. The question is formulated in accordance with the questionnaire used by World Values Survey‟s to make eventual comprehensions with other countries possible. An alternative way to ask my head question might have been to add a condition even to the inequality averse argument or to just state that we need larger income differences without any added value assessments.

Another way to formulate the issue could have been with a real world example,

relevant for the respondent. This is because people often have one view about how income

should be distributed in the theory, but when it comes to real world examples and concrete

situations, they would no longer support their theoretical view of how to best allocate income

(Amiel & Cowell, 1999a). To get an indication of to which extent the respondents is likely to

support their theoretical view even in reality I intend to test how the question about income

differences is related to the question about how much response the government should take

for the people in an economy. This is because the government is one of the two main

12

redistribution channels in the economy and a question about how this channel ought to work is less abstract than a question about just unalloyed income distribution.

The independent variables used in the regression model are all related to the theories about attitudes towards income inequality presented in section 1.2. The trust variable is an index built upon people‟s tendency to trust different groups. First, the respondents indicated on a scale from 1-5 whether they trusted a group or not. These answers were in turn classified into four different groups. One subgroup consisted of people in direct connection to the person (e.g. family and neighborhood), one with other religious groups, one with different ethnical groups and one with people from other countries than Kenya. These classes were finally unitized into a trust index and this index was included as a variable in the model. To measure the reliability of the index, a test with Kuder-Richardson Formula 20 were made. The result showed a reliability on 0.7031, which is acceptable (Appendix 2).

To measure whether Hirschman‟s theory can be used as an explanation factor for people's attitudes towards income inequality, questions has been asked about people‟s

experience of free choice and control over how the future life turns out, as well as about their satisfaction with their current situation, concerning their economic situation and their life as a whole. To test for Hirschman‟s theory I first created an index out of people‟s satisfaction with their current situation. When testing the reliability of the index, I got a value of Cronbach´s alpha on 0.8667, which is an acceptable value (Appendix 2). My intention is to create an interaction variable out of the index and the control and free choice variable. In the

regressions I will start by testing how each of the two variables impact on attitudes towards income inequality, but since Hirschman‟s theory relates people‟s current situation to their prospects about their future, I also intend to test for how this interaction variable between the satisfaction index and the control and free choice variable impact upon the attitudes towards income inequality variable. If there are indications of an interaction effect, I will use the interaction variable when testing for how gender interacts with the trust index respectively the Hirschman-related variables. To see how people‟s current income matters for the respondent‟s view about how to distribute income, the household‟s income level were asked for in the questionnaire. The reason why I asked for the household‟s income level and not for the earnings of the individual responding the questionnaire is because most people in the Kenyan society lives together in families and shares the income between the inhabitants in the

household. The household‟s gathered income is consequently more important than the individual‟s wage, but since the size of the households differs significantly between

respondents, income per person in the household is more adequate to use in the model. The

13

specified income level for a household has therefore been divided with the number of people living in the household.

One underlying assumption for regression models is that there is no presence of multicollinearity. Multicollinearity is when a correlation between the independent variables in the model is present and causes inefficient estimates. To see if the independent variables in the model are interrelated to each other, a couple of Spearman‟s correlation coefficients have been made. The result shows that there is an indication for multicollinearity affecting

variables integrated in interaction variables. This is though natural, since the same variable is included in both the correlated variables and therefore, by definition, is nearly correlated with itself. There is no fixed limit for when a correlation is seen as problematically high, but a common limit is on 0.8 and above. In the table (Appendix 2) correlations over 0.80 are bolded.

To compensate for potential omitted variable bias, different control variables has been added to the regression. These variables are: income per person, which area the respondent lives in, marital status, children, education, if the breadwinner is employed, the respondents type of employment, if the respondent is interested in politics, proud to be Kenyan, believe in god and if it thinks ethnical diversification is good for the country or not. Descriptive statistics for the variables is presented below.

The mean value for the variable attitudes towards income inequality is 6.250, with a standard deviation on 2.944. Because the depended variable is measured on an ordinal scale the mean value should not be interpreted as an absolute value, but can be used as an indication about how the respondents views the issue.

Figure 3.Distribution for the variable Attitudes towards income inequality

14

Table 1. Descriptive statistics

Attitudes towards income distribution

Percent Frequency

Gender Male 50.56 91

Female 49.44 89

15-24 22.91 41

25-34 28.49 51

Age 35-44 21.79 39

45-54 18.99 34

55- 7.82 14

Area Town 48.33 87

Rural areas 51.67 93

Married 46.11 83

Living together as married 12.22 22

Divorced/Separated 5.56 10

Widowed 8.33 15

Single 27.78 50

Children No children 32.22 58

One child or more 67.78 122

No formal education 6.67 12

Education Primary school 28.33 51

Secondary school 42.78 77

University 22.22 40

0 - 1 999 KSh 63.22 110

2 000 - 3 999 KSh 19.54 34

4 000 - 5 999 KSh 9.20 16

Income per person 6 000 - 7 999 KSh 4.60 8

8 000 - 9 999 KSh 1.15 2

10 000 - KSh 2.29 4

Breadwinner employed Employed 58.62 102

Not employed 41.38 72

Private 67.78 122

Kind of employment Public/Governmental 21.67 39

Non profit organization 10.56 19

Interest in politics Interested 39.11 70

Not interested 60.89 109

Proud to be Kenyan Proud 83.89 151

Not proud 16.11 29

Believe in god Do beleive 97.78 176

Do not beleive 2.22 4

Christian 90.96 161

Type of religion Muslim 8.47 15

Other 0.56 1

Trust no groups 21.23 38

Trust one group 15.64 28

Trust Index Trust two groups 18.44 33

Trust three groups 31.28 56

Trust all four groups 13.41 24

Mean Std. Dev.

If the respondent thinks income should be more equally distributed or not 6.25 2.94

If government should take more responsibility or not 4.82 3.03

If the respondent is satsisfied with its life or not 4.56 2.64

If the respondent is satisfied with its economic situation or not 4.07 2.35

If the respondent believe it has a free choice or not 6.80 2.50

If the respondent think ethnical diversification enriches a country or not 5.61 3.050

15

4. Econometric model

To analyze the data collected the statistical program STATA has been used. The methodology of the thesis is an ordinal logistic regression. The outcome of the ordinal logistic model will be explained in terms of odds ratios, which are built up of probabilities for different outcomes in attitudes towards income inequality. To measure the relationship between two independent variables and their impact on the dependent variable, interaction variables will be constructed.

4.1 Ordinal logistic regression

When the dependent variable in a model is dichotomous or measured on an ordinal scale, the basic assumptions for a linear regression is not valid and a logistic model should be used. An ordinal logistic regression is derived from a binary logistic regression, but is generalized to include more than two possible outcomes, which is the case in a logistic regression with a binary depending variable. To explain an ordinal logistic regression model, I will start by describing a logistic regression model.

A logistic regression model investigates the correlation between the explaining variable and the probability that the dependent variable will adopt a value:

Prob(Y=1,0) = f(x).

The result from the regression is presented in shape of odds ratios, which is the ratio of the probability that an event will occur, divided by the probability that the event will not occur.

The basic logistic regression is specified by:

Logit(P

) = ln(P

/ (1-P

)) = β

+ β

X

+…+ β

X

Where i = 1,….,n symbolize each individual in the sample, P is the probability and (P

/ (1- P

)) is the term for the odds ratio. The basic logistic regression is expressed in terms of logarithmic odds ratio, but can be simplified by taking the natural logarithm of the function:

P

= Prob(Y

=1│X

) = 1/(1+e

)

In an ordinal logistic regression the focus of interest is the odds that the independent variable

attains a higher value. That is

16

Ѳ

= prob(score ≤ j) / prob(score > j)

where j is the possible values on the dependent variable. The last category included in the independent variable do not have an odds associated with it since the probability of attaining and including the last category is 1.

Odds ratios show the relative change in the odds when the independent variable changes one unit and the remaining, independent variables are held constant. The odds ratios are based on the odds for the dependent variable to attain a higher category rather than to attain a lower one. It can be interpreted in the following way: Odds ratios lower than one signifies a reduced likelihood for the event to occur compared to the reference group, ceretis paribus. In the same way, an odds ratio higher than one indicates a higher risk or chance for the event to occur compared to the reference group.

4.2 Interaction effects

In multiple regression analysis the model usually follows an additive form, implying that the variables in the model have independent effects on the dependent variable:

Y = α + β1x1 + β2x2 + e

However, if we suspect the effect of one independent variable to depend on, or impact upon, the relation between another independent variable and the dependent variable, a multiplicative term is more useful:

Y = α + β1x1 + β2x2 + β3 x1x2 + e

In a regression model with an interaction effect, the slope of the coefficient

measures the

interaction effect of the variables. The relation between the two independent variables and the

dependent one is illustrated graphically below.

17

Figure 2. The variable A impact on attitudes towards income differences. Variable B is here placed as an interaction variable. That is, if variable B indicates an impact on both variable A and the dependent variable, it is useful to integrate an interaction variable between variable A and B in the model.

5. Result

Below the results of the ordinal logistic regression with attitudes towards income inequality as the dependent variable will be presented. Column one shows a regression model without interaction variables, column two presents a regression model with an interaction variable between the satisfaction index and the free choice variable, column three shows a regression models with an interaction effects between gender and trust, and column four shows a regression model with interaction variables between gender and the satisfaction index and between gender and the free choice and control variable added. Descriptive data for the variables is presented in section 2.3.

Table 2. Ordinal logistic regression presented in odds ratios and with attitudes towards income inequality as a dependent variable. Standard deviations are assigned within brackets under the odds ratio.

(1) (2) (3) (4)

VARIABLES Attitudes towards

income inequality

An interaction variable between

satisfaction and free choice added

An interaction variable between

gender and trust added

An interaction variable between

gender and free choice and gender

and satisfaction added

Male Ref(1) Ref(1) Ref(1) Ref(1)

Female 2.571*** 2.415** 2.016 1.068

(.884) (0.839) (1.127) (1.168)

Trust index 0.014 0.999 0.956 1.003

(0.123) (0.122) (0.155) (0.122)

Gender*Trust Index 1.130

(0.250)

18

Satisfaction index 0.909*** 1.014** 0.909*** 0.915**

(0.031) (0.095) (0.031) (0.040)

Control and Free choice 1.162** 1.319** 1.157** 1.091

(0.075) (0.160) (0.075) (0.153)

Satisfaction index*Free Choice 0.985

(0.012)

Gender *Satisfaction index 1.000

(0.067)

Gender *Free choice 1.143

(0.153)

Income per person 0.000 0.000 0.000 0.000

(0.000) (0.000) (0.000) (0.000)

Age 0.985 0.984 0.986 0.987

(0.015) (0.015) (0.015) (0.015)

Rural areas Ref(1) Ref(1) Ref(1) Ref(1)

Voi town 1.248 1.260 1.264 1.302

(0.540) (0.542) (0.548) (0.564)

Married Ref(1) Ref(1) Ref(1) Ref(1)

Living as Married 0.995 0.995 0.966 0.964

(0.503) (0.503) (0.491) (0.487)

Divorced 5.618** 5.716** 5.742** 6.040**

(3.953) (4.041) (4.043) (4.340)

Widowed 2.245 2.490 2.267 2.451

(1.267) (1.437) (1.276) (1.426)

Single 1.290 1.317 1.310 1.343

(0.818) (0.834) (0.833) (0.854)

No Children Ref(1) Ref(1) Ref(1) Ref(1)

One child or more 2.403* 2.670* 2.366 2.247

(1.261) (0.535) (1.242) (1.187)

Education 1.002 1.014 0.010 0.970

(0.246) (0.248) (0.248) (0.241)

Breadwinner Employed Ref(1) Ref(1) Ref(1) Ref(1)

Breadwinner Unemployed 1.680 1.715 1.695 1.690

(0.585) (0.600) (0.591) (0.591)

Private Ref(1) Ref(1) Ref(1) Ref(1)

Public 0.771 0.715 0.784 0.783

(0.403) (0.378) (0.411) (0.413)

19

NGO 1.252 1.219 1.224 1.198

(0.635) (0.620) (0.622) (0.611)

Interested in Politics Ref(1) Ref(1) Ref(1) Ref(1)

Not interested in Politics 1.481 1.575 1.461 1.450

(0.465) (0.500) (0.411) (0.457)

Proud to be Kenyan Ref(1) Ref(1) Ref(1) Ref(1)

Not proud to be Kenyan 0.488 0.479* 0.493 0.518

(0.214) (0.207) (0.216) (0.225)

Do believe in God Ref(1) Ref(1) Ref(1) Ref(1)

Do not believe in God 7.834* 4.879 7.943* 9.803**

(8.619) (0.207) (0.216) (11.141)

Christian Ref(1) Ref(1) Ref(1) Ref(1)

Muslim 0.882 0.896 0.873 0.874

(0.474) (0.486) (0.471) (0.472)

Other 5.539 6.046 5.755 5.772

(8.927) (9.754) (9.290) (9.295)

Ethnical diversification 1.256*** 1.256*** 1.256*** 1.238***

(0.067) (0.067) (0.067) (0.068)

cut1 0.180

(1.232)

0.815 (1.469)

-0.248 (1.237)

-0.066 (1.335)

cut2 0.430 1.427 0.364 -0.047

(1.220) (1.461) (1.225) (1.322)

cut3 1.016 2.022 0.951 0.539

(1.216) (1.462) (1.220) (1.318)

cut4 1.486

(1.216)

2.499 (1.466)

1.421 (1.220)

1.005 (1.319)

cut5 2.313 3.332 2.250 1.830

(1.227) (1.479) (1.231) (1.329)

cut6 2.776 3.795 2.713 2.296

(1.234) (1.485) (1.238) (1.333)

cut7 3.273 4.289 3.209 2.799

(1.238) (1.488) (1.242) (1.335)

cut8 3.984 5.002 3.921 3.518

(1.247) (1.498) (1.251) (1.342)

cut9 4.388 5.409 4.327 3.925

(1.255) (1.506) (1.259) (1.350)

Observations 164 164 164 164

Standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

In the first regression model three of the investigated variables are significant on a ten percent

level. These variables are the gender variable, the satisfaction index and the control- and free

20

choice variable. The regression model shows that the female gender has a more positive view on income inequality compared to the reference group men. The female group show a 2.57 times higher odds to be in a higher category regarding attitudes towards income inequality than the male group. The satisfaction index indicates that people who are relatively more satisfied with their life situation tend to think there should be less income inequality. The odds for a lower category on the dependent variable are 9 percent smaller for every step closer to the apprehension completely satisfied. The variable control and free choice indicates that people who think they have a relatively bigger amount of impact upon how their life turns out are more likely to think there is more room for income inequality. The odds ratio shows a 16.2 percent higher odds to be in a higher category on the variable attitudes towards income

inequality for each step closer to the apprehension completely control and free choice over how the future turns out. The trust index variable and the variable income per person are not significant in the model. From the controlling variables I got four significant outcomes: The group divorced respondents shows a 5.62 times higher odds to be in a higher category compared to the reference group married. The variable children is also significant on a ten percent level and indicates that respondents with one child or more are more likely to be more positive to income differences than the reference group people without children. The odds ratio for the variable children is 2.40, which indicates that respondents with children have a 2.4 times higher odds to be in a higher category compared to the group people without children. The variable about if the respondent believes in god or not is significant on a ten percent level. It shows that people who not believe in god are 7.83 times more likely to be in a higher category on the attitudes towards income inequality variable compared to those who believe in god. The variable ethnical diversification is significant and shows a correlation between positive attitudes towards ethnical diversification and positive attitudes towards income inequality. One step closer the statement that ethnical diversification enriches life gives 25.6 percent higher odds to be in a higher category on the attitudes towards income inequality variable.

Since regression number two shows no evidence for the assumption that an interaction variable between the satisfaction index and the variable control and free choice impacts on people‟s attitudes towards income inequality, I will test the interaction effects between gender and the Hirschman related variables separately and not with an interaction variable between gender and the future expectation interaction variable.

The result for regression number three, with an interaction variable between gender

and trust, shows no indications for the assumption that an interaction variable between gender

21

and trust contribute to explain people‟s attitudes towards income inequality. This is however not surprising since the trust index variable was not significant either in the first regression.

The disappeared significance for the variable gender in the model is likely to depend upon the reported prevalence of multicollinearity between the gender variable and the interaction variable between gender and trust. The rest of the regression model shows the same result as the first regression model except for the variable children, which is no longer significant on a ten percent level.

The interaction variables between the gender variable and the satisfaction index and between the gender variable and the control and free choice variable in regression number four is not significant in the model. Even here, just as in the model with an interaction variable between gender and trust, the gender variable shows a disappeared significance, which is likely to depend upon the reported prevalence of multicollinearity between the gender variable and the interaction variables in the model. Even the variable control and free choice shows a disappeared significance. The controlling variables, significant in the first regression model, are still significant except from the variable children.

Logistic regression is often seen as a factor oriented method, where the main focus is to test the effect of different single variables. The amount of explained variation in the model is therefore of an inferior meaning. Possibly is this an explanation for the absence of distinct standards on how to evaluate a whole model made with logistic regression (Hagquist &

Stenbeck, 1998). I decided to use the likelihood ratio, the p-value associated to the chi-square and the pseudo-R

as indicators for the validity of my model. The p-value, built upon the likelihood ratio, shows that the model as a whole is statistically significant (p = 0.000). Since the R

measure is not pertinent in logistic regressions the pseudo-R

is used as an indicator for the explanation grade of the model. It cannot be interpreted in the same way as the average determination coefficient; instead it should be expounded as an approximation of the R

value. The values for the pseudo R

in my regression models varies between 0.0917 and 0.0938.

6. The state as a redistributional institution

As earlier stated, the government together with the market is seen as the main redistributional institution in the economy. In the questionnaire, a question about governmental responsibility was included and asked in the following way:

What do you think about the following issue? Place it on the scale below.

22

The government should take more responsibility People should take more

to ensure that everyone is provided for responsibility to provide for themselves 1 2 3 4 5 6 7 8 9 10

To see if there is any indications for the suspicion that the respondents theoretical view not necessarily correspond to their view about how to allocate the income in a real world situation, a regression model with the question about governmental responsibility as the dependent variable is made. I will also see how the variable governmental responsibility is correlated with the variable attitudes towards income inequality. If the theoretical view corresponds to the real world example there should, with all probability, be a correlation between these two variables. Certainly, an absent correlation might depend on other factors than a missing link between the theoretical and the practical view, e.g. the respondents might think that the market is much more suitable than the government to redistribute income through.

The mean value of the governmental responsibility variable is 4.822 and the standard deviation is 3.025. This is comparable to the values on the variable attitudes toward income inequality, which had a mean value on 6.250 and a standard deviation on 2.944. Just as before the mean values should be used as indication since the variables is measured on an ordinal scale. When testing the correlation between attitudes towards income inequality and

governmental response Spearman‟s roh attained a value at 0.1034 (appendix 2), which is not high enough to establish a correlation between these two variables. Below, an ordinal logistic regression model has been made with governmental responsibility as the depending variable.

The explaining variables are the same as in the regression made with attitudes towards income inequality as the depending variable.

Table 3. Ordinal logistic regression presented in odds ratios with governmental responsibility as an independent variable. Standard deviations are assigned within brackets under the odds ratio.

(1)

VARIABLES How much responsibility

the government should take

Male Ref(1)

Female 1.377

(0.458)

Trust Index 0.841

(0.104)

23

Satisfaction index 1.044

(0.036)

Control and free choice 0.941

(0.058)

Income per person 0.999*

(0.000)

Age 1.008

(0.015)

Rural areas Ref(1)

Voi town 1.573

(0.681)

Married Ref(1)

Living as Married 0.870

(0.436)

Divorced 3.829*

(2.691)

Widowed 0.805

(0.416)

Single 1.392

(0.922)

No children Ref(1)

One child or more 0.907

(0.508)

Education 1.287

(0.313)

Breadwinner Employed Ref(1)

Breadwinner Unemployed 1.014

(0.342)

Private Ref(1)

Public 0.015

(0.499)

NGO 0.495

(0.256)

Interested in Politics Ref(1)

Not Interested in Politics 1.051 (0.329)

Proud to be Kenyan Ref(1)

Not Proud to be Kenyan 1.160

(0.483)

Believe in God Ref(1)

Do not Believe in God 1.317

(1.478)

24

Christian Ref(1)

Muslim 2.250

(1.237)

Other 0.929

(1.493)

Ethnic diversification 1.030

(0.053)

cut1 -0.093

(1.178)

cut2 0.166

(1.177)

cut3 0.496

(1.177)

cut4 0.720

(1.178)

cut5 1.741

(1.187)

cut6 1.893

(1.190)

cut7 2.264

(1.197)

cut8 2.855

(1.209)

cut9 3.437

(1.220)

Observations 164

In the regression with governmental responsibility as the dependent variable two of the variables are significant on a ten percent level, one explaining variable and one controlling variable, namely the variables income per person and if the respondent is divorced. These result shows that respondents with relatively less income per person shows a 1 percent higher tendency to be in a lower category, closer to the statement that the government should take more responsibility to see that everyone in the society is provided for. The other significant variable, divorced, shows a 3.83 times higher odds to be in a higher category compared to the reference group married.

However the p-value for the likelihood ratio is not significant on a ten percent

significance level (p=0.469), which indicates that the model as a whole is not significant and results from the regression should be interpreted with caution.

7. Conclusions

This paper attempted to evaluate how attitudes towards income differences depend on gender,

trust, future expectations and current income level. Given the theoretical question “how do

you think income should be distributed?” the respondent indicated, on a scale from one to ten,

25

if income should be more or less equally distributed. Below, the outcome of the regressions is discussed, the expected as well as the unexpected results. I attempt to draw conclusions out from the result, which in some aspects might be difficult, especially for the unexpected outcomes, which I in some cases have no good explanations for.

The odds ratio for the independent variable gender has not turned out as expected; the result shows that the female group thinks it is more acceptable with a higher amount of income inequality than the reference group male does. The other significant variables, which are satisfaction and amount of control and free choice, go partly in the expected direction.

That is, people with a more positive experience of free choice and control tend to represent a more positive view towards income inequality. The trust index variable and the variable income per person are not significant in the model. Neither are the interaction variables used in the model significant. The circumstance, that the income per person variable is not

significant in the regression model, gives an indication for the accuracy in assuming that there are other, more important factors than current income level to consider when explaining people‟s attitudes towards income inequality.

As said, I expected the female group to be more inequality averse than the male group, but this was not the case. A partially explanation for this result might be related to the control variable children. According to Marinda (2006), the female group in Kenya takes more responsibility than males to see that the children in the household are provided for and she stresses that children in female headed households is better off concerning their nutrition status. Since the variable children showed that respondents who have one child or more also shows a bigger tendency to think it is acceptable with more income inequality, these questions might be related. Why the results goes in the direction it does is though unclear. Perhaps do people who need to take a greater amount of responsibility, either for their own situation, like divorced or widowed, or for another person, like female respondents do for the children (Marinda, 2006), think it is more acceptable with a higher amount of income inequality. This is however not a conclusion which can be empirically reinforced in this thesis.

The outcome for the control and free choice variable was expected with regard to

Hirschman‟s tunnel effect theory. The satisfaction index is also significant in the model, but

the control and free choice variable is a better indicator for Hirschman‟s tunnel theory since

the variable actually gives an indication about how the respondent views its future and not

only its current situation. However, Hirschman theory attach people‟s current situation to their

future expectations and to be able to really say anything about how the data connects to

Hirschman‟s theory, the respondent views about its future should be anchored in the person‟s

26

current situation. This is not the case here, since the interaction variable between the satisfaction index and the control and future expectation variable was not significant in the regression model. Therefore, it might be relevant to think that Hirschman‟s tunnel effect theory can constitute a partial explanation for people‟s attitudes towards income inequality, but not solely and not unchallenged.

Another significant result, which I have no good explanations for, is that people who assigned that they believe in god shows a lower tendency to think it is acceptable with income inequality compared to the group people who do not believe in god. The same is to be said about the variable ethnic diversification, which shows a positive correlation between a positive view on ethnical diversification and positive attitudes towards income inequality.

This is however an interesting result since Kenya is an ethnical segregated country where almost all political power (and thereby also income allocation) is based upon ethnical appurtenant. (SID, 2006)

As mentioned in the beginning of this essay, there is supposed to be a correlation between the view about how to distribute income and the size of the cake to distribute. In Kenya, the size of the cake differs significantly between provinces, which are likely to impact on the possibility to extend the validity of the result to other places in the country. It is hard to say if this study made in Voi division gets any external validity for the rest of Kenya even from other aspects. This is because the Kenyan society differs a lot between provinces, as well as between towns and rural areas, when it comes to values regarding, among others, the female sex. There are also substantial differences between different provinces in the country when it comes to factors related to inequality, such as poverty, current income distribution and access to different functions, e.g. health care and education.

The explanation rates for the regression models are not very high and there is likely to

be other, external variables which impacts on how people develop their view about the best

way to distribute income. Personal attributes cannot solely be an explanation factor for

attitudes towards income inequality, the question need to be put in its societal context. How

the society is built up is likely to impact on how people develop their attitudes towards

income inequality. For example, I expect current, as well as earlier political system and

current inequality level in the country to constitute explanation factors for how people

develop their attitudes towards income inequality.

27

8. References

Amiel, Y. & Cowell, F.A (1999a), Thinking about inequality – Personal judgment and income distributions, Cambridge University Press, Cambridge.

Amiel, Y., Creedy, J. & Hurn, S. (1999b), „Measuring attitudes towards inequality‟, Scandinavian Journal of Economics 101, 83-96.

Ben-Ner, A., Putterman, L., 2001. „Trusting and trustworthiness‟, Boston University Law Review 81, 523-551.

Bjørnskov, Christian (2007), “Determinants of generalized trust: A cross-country comparison” in Public Choice 130 (1–2), 1–21.

Boix,

Cowell, F.A & Cruces, G. (2004) “Perceptions of inequality and risk” in Studies on Economic Well-Being: Essays in the Honor of John P. Formby, Emerald Group Publishing Limited, Bingley.

Dahmström, K. (2005), Från datainsamling till rapport: att göra en statistisk undersökning, Studentlitteratur, Lund.

Davidovitz, L. & Kroll, Y. (2004), “On the attitude towards inequality” in Inequality, welfare and income distribution, Emerald Group Publishing Limited, Bingley.

Djurfeldt, G. & Barmark, M. (2009), Statistisk verktygslåda – multivariat analys.

Studentlitteratur AB, Lund.

Eckel, C. C. & Wilson, R. K. (2004), “Is trust a risky decision?”, Journal of Economic Behaviour & Organization 55, 447-465

Ellis, A. et al. (2007), Gender and economic growth in Kenya – Unleashing the power of women, World Bank, Washington D.C.

Knack, S. & Keefer, P. (1997), “Does social capital have an economic pay-off? A cross-country investigation” in Quarterly Journal of Economics 112 (4), 1251–1288.

Hagquist, C. & Stenbeck, M. (1998), “Goodness of Fit in Regression Analysis – R

and G

Reconsidered” Quality and Quantity 32 (3)

Harkness, J. (2003), “Questionnaire translation” in Cross-Cultural Survey Methods (Wiley Series in Survey Methodology), John Wiley & Sons, Inc., New Jersey.

Hirschman, A.O. (1973), „The changing tolerance for income inequality in the course of

economic development, with a mathematical appendix by Michael Rothschild‟, Quarterly

Journal of Economics 87, 544–56.

28

Marinda, P. (2006), Effects of Gender Inequality in Resource Ownership and Access on Household Welfare and Food Security in Kenya, Peter Lang Europäischer Verlag der Wissenschaften, Frankfurt am Main.

Morgan, W., Katz, M. & Rosen, H. (2009), Microeconomics, McGraw-Hill Higher Education, Berkshire.

Perkins, D. H., Radelet, S. & Lindauer, D. L. (2006), Economics of development 6

ed., W.

W. Norton & Company, London.

Republic of Kenya (2009), District development plan 2008-2012, National Development and Vision 2030, Nairobi.

Sen, Amartya K. (1985), Commodities and Capabilities, Oxford University Press, Oxford.

Society for International Development (SID) (2006), Readings on inequality in Kenya – Sectoral dynamics and perspectives, Nairobi.

Szirmai, A. (1988), Inequality observed - A Study of Attitudes Towards Income Inequality, Gower Publishing Company Limited, Aldershot.

United Nation Development Program (UNDP) (2009), Human Development Report 2009.

Overcoming barriers: Human mobility and development, New York.

Vickrey, W. (1945) „Measuring marginal utility by reactions to risk‟ Econometrica 13:4 319- 333.

World Values Survey (2011), http://www.worldvaluessurvey.org/ (2011-12-05).

Zak, P.J. & Knack, S. (2001), “Trust and growth” in Economic Journal 111 (470), 295-321

29

Appendix 1

Hello! My name is Maria Carlsson and I am a student at the Department of economics at Uppsala University in Sweden. I am carrying out a study about what people value in life and I would like to ask about your view on a number of different subjects. Please try to answer all questions and feel free to ask me to specify if anything seems to be unclear. My phone number is 071-7736978. Of course your input will be treated strictly confidential and you are free to stop your contribution whenever you want to. Please tick in the box or circle in the answer that suits you the best.

Thank you for your participation!

Questionnaire

1. Gender:

Male Female

2. Age:

15 – 24 45 – 54

25 – 34 55 –

35 – 44

3. Are you currently

Married Separated

Living together as married Widowed

Divorced Single

4. Do you have any children?

No children Four children

One child Five children

Two children Six children

Three children Seven children or more

5. How many persons live in your household?

1 person 6 persons

2 persons 7 persons

3 persons 8 persons

4 persons 9 persons

5 persons more than 9 persons

6. Would you say you are

A religious person Not a religious person An atheist

7. Which religion or religious denomination do you belong to?

30

Catholic Shia Muslim

Protestant Hindu

Orthodox Sikh

Jew Local religion

Sunni Muslim Other (please specify):________________

8. What is the highest education level that you have attained?

No formal education Incomplete primary school Complete primary school Incomplete secondary school Complete secondary school

Some university-level education, without degree University-level education, with degree

9. Are you employed now or not? If yes, about how many hours a week? If more than one job, only the main job.

Yes, paid employment

Full time employment (<30 h/week) Part time employment (>30 h/week) Self employed

No, no paid employment

Retired/Pensioned

Housewife, not otherwise employed Student

Unemployed

10. In which profession/occupation are you doing most of your work? If you are not currently employed, characterize your major work in the past (please answer on the line below):

11. Are you working for the government or public institution, for private business or industry, or for a private non-profit organization? If you do not work currently, characterize your major work in the past! Do you or did you work for:

Government or public institution Private business or industry Private non-profit organization

12. Approximately how much does your household earn each month?

31

Less than 3 000 KSh 3 000 KSh – 5 999 KSh 6 000 KSh – 8 999 KSh 9 000 KSh – 11 999 KSh 12 000 KSh – 14 999 KSh 15 000 KSh – 18 999 KSh 19 000 KSh – 21 999 KSh More than 22 000

13. Are you the chief wage earner in your household?

Yes No

14. If not, is the chief wage earner of your household employed now or not?

Yes No

15. How satisfied are you with the financial situation of your household? Place on the scale below, where one is completely dissatisfied and ten is completely satisfied. Please circle in the answer that suits you best.

Completely dissatisfied Completely satisfied

1 2 3 4 5 6 7 8 9 10

16. How satisfied are you with your life as a whole these days? Place on the scale below.

Completely dissatisfied Completely satisfied

1 2 3 4 5 6 7 8 9 10

17. For each of the following, indicate how important it is in your life. Please tick in the box that suits you the best.

Very important Rather important Not very important Not at all important Family

Friends Leisure time Politics Work Religion

18. How often do you meet people from different religions, ethnical groups and different countries? Please tick in the box that suits you best.

32

Every day Once a week Once in a Once in a Less than once

month year in a year

a. From another religion

b. From another ethnic group

c. From other African countries

d. From non African countries

19. How interested would you say you are in politics? Are you Very interested

Somewhat interested Not very interested Not at all interested

20. Some people feel they have completely free choice and control over their lives, while other people feel that what they do has no real effect on what happens to them. Please use this scale to indicate how much freedom of choice and control you feel you have over the way your life turns out:

No choice at all A great deal of choice

1 2 3 4 5 6 7 8 9 10

21. Would you say that most people can be trusted or that you need to be very careful in dealing with people?

Most people can be trusted Need to be very careful

22. I would like to ask you how much you trust people from various groups. Could you tell me for each whether you trust people from this group completely, somewhat, not very much or not at all? For every statement, please tick in the box that suits you best.

Trust Trust Do not trust Do not trust completely somewhat very much at all a. Your family

b. Your neighborhood

c. People you know personally d. People you meet for the first time e. People from another ethnic group

f. People of another nationality i. from a neighboring country ii. from an another African country iii. from an non-African country g. People of another religion

i. Christians

33

ii. Muslims iii. Hindus iv. Sikhs v. Local religions vi. Jews

23. How proud are you to be Kenyan?

Very proud Quite proud Not very proud Not at all proud I am not Kenyan

24. About ethnic diversity, with which of the follow views do you agree? Please use the scale to indicate your position:

Ethnic diversity erodes a country‟s unity Ethnic diversify enriches life

1 2 3 4 5 6 7 8 9 10

25. How about people from other countries coming here to work. What do you think that the government should do?

Let anyone come who wants to

Let people come as long as there are jobs available

Place strict limits on the number of foreigners who can come here Prohibit people coming her from other countries

26. For each of the following statements, can you tell me how strongly you agree or disagree with each. Do you strongly agree, agree, disagree, or strongly disagree? Please tick in the answer that suits you best.

Strongly Agree Neither agree Disagree Strongly agree nor disagree disagree Being a housewife is just as fulfilling as working for pay.

On the whole, men make better political leaders than women do.

A university education is more important for a boy than for a girl.

To fully develop your talents, you need to have a job.

It is humiliating to receive money without working for it.

People who don‟t work become lazy.

34

Work is a duty toward society.

Work should come first, even if it means less free time.

27. What do you think about the following issues? Place it on the scale below.

Income should be made more equal We need larger income differences as

incentives for individual effort 1 2 3 4 5 6 7 8 9 10

The government should take more responsibility People should take more

to ensure that everyone is provided for responsibility to provide for themselves 1 2 3 4 5 6 7 8 9 10

35

Appendix 2

Table 4. Trust index evaluated with Kuder-Richardson Formula 20

Average interitem of covariance 0.0817668

Number of items in the scale 4

Scale reliability coefficient 0.7031

Table 5. Satisfaction index evaluated with Kronbach´s alpha

Average interitem of covariance 4.78958

Number of items in the scale 2

Scale reliability coefficient 0.8667

Table 6. Correlation between attitudes towards income inequality and governmental responsibility

Number of observations 180

Spearman‟s Roh 0.1034

Test of H_o: Attitudes and State are independent

Prob >

│t│