Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
5. OUTREACH OF MICRO FINANCE PROGRAMMES AND TARGETING OF
THE POOR
Chapter 5
5. OUTREACH OF MICROFINANCE PROGRAMMES AND
TARGETING OF THE POOR
5.1 INTRODUCTION
A number of anti-poverty policies and programmes have been launched in India
during the 1970s and 1980s, including those of credit delivery programmes, targeting
the poor. Various methodologies have been devised and nation wide surveys have
been conducted in identifying the poor for the successful delivery of the programmes.
Unfortunately, such programmes have always been under criticism for mistargeting
either due to lackluster implementation or elite capture at the ground level. For
example, Gaiha eta!. (2001a) argued that benefit to the rural poor of two major anti
poverty programmes (Rural Public Works and Integrated Rural Development
Programme) are likely to be limited, given their gross mistargeting. Larger sections
among the poor were not covered and moreover, the non-poor were the majority
among the participants. Similar outcomes were also reported by Dreze (1990) in his
study on IRDP in the state of Uttar Pradesh. Such mediocre targeting and performance
of the public programmes on poverty alleviation during the 1980s questioned the
poverty alleviation strategies, particularly identifying poor households/individuals,
besides other bottlenecks in governance and appropriateness. During this period,
many developing countries of Asia, Africa and Latin America reported success of a
new form of targeting the poor which was pioneered by Grameen Bank in Bangladesh.
Inspired by popularity of Grameen's group-based targeting mechanism, and
recognising credit as the main instrument in attacking poverty, microfinance
programmes in India are promoted primarily as a substitute of anti-poverty
programmes through delivering credit for expanding production and employment
choices of poor.
The country-wide spread of microfinance programmes in recent years is
overwhelming and it has generated concern among researchers and policymakers in
scrutinizing the outcomes of such large outlay, most particularly the issue of targeting.
Indeed, the concern arises while one looks at the people in the lower strata of income
distribution -the very poor category. They may be too poorly informed, educated or
nourished to take advantage of the programme, they may not possess required
Chapter 5
documents such as caste certificates or proofs of residence, they may be socially
ostracized or they may even face agency problems that lead bureaucrats and local
politicians to divert resources to the non-target sections of population. Further, there is
empirical evidence from a variety of social programmes in both developed and
developing countries that information sets differ, even among those eligible, and that
participation rates vary widely and are sensitive to programme design. In Bangladesh,
programmes with clear eligibility conditions have often been violated them due to
expectation of higher returns from wealthy borrowers (Morduch, 1998; Rahman,
1999). In the case oflndia, many ofthe group-based microfinance programmes do not
have any clear cut guidelines in targeting and therefore, the possible bias against the
very poor cannot be ruled out. One of the leading microfinance institutions (MFis) in
India believes that MFis are designed to serve the poor and, as a trickle down effect in
the rural economy, the poor will take care of the very poor (Mahajan, BASIX
webpage). Although there has been a plethora of research in assessing the impact of
these programmes on a variety of outcomes 1, in developing as well as developed
countries, there has been little research on how well targeted these group-based
microcredit programmes actually are. Besides, researchers also widely differ in
methodology of identifying poor. In the case of India, both the issues are still under
scrutiny and there is little empirical research from academia in identifying the
targeting of microfinance programmes.
The chapter, thus, examines the extent to which these new institutional arrangements
have been successful in delivering financial services to the poor or the very poor. In
view of unavailability of published data for measurement of targeting, the present
study employs household level data collected from field work in Madhya Pradesh.
The first section of the chapter briefly discusses the review of literature pertaining to
the issue of targeting of microfinance programme. The second section discusses the
data and methodology of the study. The third section analyses the estimation results.
5.2 LITERATURE REVIEW
Microfinance programmes have been under scrutiny, since their inception,
particularly on the issue of targeting. Basically targeting has been debated as a
1 See Hulme and Mosley, 1996a; 1996b; Pitt and Khandker, 1996; 1998; Morduch, 1998; Zaman, 2000; Puhazhedi and Satyasai, 2000; Puhazhedi and Badatya, 2002, Khandker, 2003.
[i44]
Chapter 5
methodological problem. Measuring poverty outreach of clients of microfinance
programmes by using any particular methodology is also difficult because the profile
of clients and concept of poverty varies across regions and population groups. So
poverty measured in a purist fashion (e.g. those living below x number of rupees) may
not be a very good indicator for targeting, because measuring income of household is
itself difficult and importantly, poverty has been looked at through the window of
capability of the household rather than simple income (Henry et al., 2000). Further,
unavailability of income data and questions on its reliability have very often
motivated the practitioners of microfincance programmes to use alternate indicators of
targeting, such as SEF's Participatory Wealth Ranking, Cashpor House Index and
CGAP's proposed poverty assessment tool which help in deriving relative poverty
instead of absolute income poverty. Navajas, et al. (2000) argued that unless there is
an appropriate targeting tool, the poorest will either be missed or they will tend to
exclude themselves because they do not see the programmes as being for them.
Besides methodological problems, there are design issues of microfinance
programmes which also affect targeting. For instance, if the designed products of the
microfinance programme do not suit the need of the poor, then their proportion
diminishes over time (Wright, 2000).
Many empirical studies have attempted measuring targeting of microfinance
programmes by employing different methodologies of measuring poverty 2 .
Constructing an index of fulfillment of basic needs, Navajas et al. (2000) found that
most of the poor households reached by the microfinance organizations were near the
poverty line. They were the richest of the poor. Amin et a!. (2003) used distribution
functions of consumption expenditure between members and non-members to
examine stochastic dominance of one distribution over the other. Further, they
attempted to measure vulnerability in a risk sharing framework of Townsend (1994)
and conclude that microcredit is successful at reaching the poor but it is unsuccessful
in reaching the vulnerable. Some studies have attempted measuring targeting of
microfinance programmes by estimating the differences between the share of
participants and non-participants in some pre-determined intervals of the income
distribution, say income quartiles, quintiles or deciles. This is especially common in
studies concerned with the incidence of public spending on infrastructural facilities
2 See Navajas eta!., 2000; Am in eta/., 2003; Diagne and Zeller, 2001.
Chapter 5
(van de Walle and Nead, 1995; Castro-Leal et al., 1999). With no targeting, the share
of an income group in the benefits of public spending would be equal to its population
share and in well targeted programmes the share of benefits to the poor is high
relative to their population share. Diagne and Zeller (2001) have examined targeting
by using probability regressions which estimates probability that a household
participates in the programme as a function of income and other household
characteristics. The use of probability regression takes into account the nonlinearities
in the marginal effects of income on participation. However, the method has been
subject to criticism as it assumes that the probability of participation is monotonic in
each of the explanatory variables. So some studies have preferred non-parametric
estimates. For example, if the set of households with low participation rates is small
relative to the population, they may also remain undetected when examining the
proportion of programme participants in arbitrarily chosen intervals of the income
distribution, especially ifthey are in income groups where other households have high
participation rates. In a study of microfinance programme in India, Somanathan and
Dewan (2003) used empirical distribution functions to show that there is possibility of
the very poor to get excluded from the programme. Ravall ion ( 1991) had used
empirical distribution functions to evaluate targeting in the Employment Guarantee
Scheme in Maharashtra and, similarly, Gaiha (2001b) used the distribution functions
to evaluate the targeting of anti-poverty programmes in India.
However, there is increasing acceptance that traditional Microfinance programmes are
not reaching the poorest of the poor-indeed they are rarely reaching the bottom 10-
15 percent of the population. This is largely driven by the nature of the financial
services provided by the MFis which force the poor members of the community to
choose not to join, and those who are required to guarantee or follow-up their loans to
chose to exclude the poorest. An MFI's ability to attract the poorest depends on the
financial services it offers and whether they have been designed to be appropriate for
the needs of the poor members ofthe community. Rutherford (1995) and Wright eta/.
(1997) argued that the exclusion of the poorest is also driven by the emphasis on
credit delivery by many organizations. For the poorest households the opportunities
for productive use of loans are limited and the risk of taking loans that are repayable
on a weekly basis are unacceptably high. Similarly, Gaiha (2001 b) reviewed the
Maharashtra Rural Credit Project (MRCP), a microcredit scheme in India, and draws
Chapter 5
attention to the deficiencies in the design and implementation ofthis scheme that limit
the participation of the poorest and the benefits accruing to them. In a study of SEWA
Bank in Ahmedabad, it has been found about 40 percent of borrower households live
in poverti. Further, the sensitivity ofthe poverty headcount suggests that households
in the sample cluster around the poverty line while most ofthem are either just above
it or just below it. So, clearly it does not include the poorest (Chen and Snodgrass,
1999).
An interesting study of ASA ( 1997), surveying 626 respondents, revealed various
reasons for which the poorest do not prefer to join the group. Lack of minimum
clothing to attend meetings, infl uence of the local elite (moneylenders, religious
leaders, union chairman etc.) to not to join the programme, the screening procedure of
the programme to exclude the risky clients and the obligation of weekly savings were
some of the reasons which excluded the poorest from the programme.
In view of the above we propose to analyse targeting by use of different
methodologies, including means tests, non-parametric tests, probability regressions
and non-parametric tests using distribution functions of income as also level of living
index which is constructed by multivariate analysis. The use of different
methodologies will help us to arrive at a conclusion which may be more credible and
acceptable to the larger audience.
5.3 DATA AND METHODOLOGY
5.3.1 Data
As mentioned in chapter 3, published data sources are insufficient to measure
targeting of microfinance programmes. So, we collected household level data from
the fieldwork carried out in two different districts of Madhya Pradesh. The study was
focused on three most popular microfinance programmes in India, as they are
implemented. The detail description of the study area, microfinance programmes,
survey methodology and sample selection is provided in chapter 4. However, our total
sample consists of 270 households, with participants and non-participants sample
being in the ratio of 2: 1.
3 The study uses poverty line specified by the World Bank and adjusts it for purchasing power parity (PPP) and prices.
Chapter 5
5.3.2 Methodology
5.3.2.1 Difference between Population Groups- Parametric Approach
A general method of observing absolute differences in participant and non-participant
households is to compare the average value of various indicators from the two
samples by using an appropriate test statistic. For example, we apply t-test to test·
whether the mean for per capita income is the same for participants and non
participants. It may be noted that the t-test assumes that the samples are drawn
randomly from two normal populations with the same unknown variance. In case the
population distributions are not normal, but the sample size is fairly large, one can
still apply t-test by invoking central limit theorem which treats the distribution to be
asymptotic normal. If the number of groups exceeds two, we apply ANOVA test to
measure whether the mean indicator is same across the groups.
A number of indicators are compared for participants and non-participants by using t
test procedure. It has been noted earlier that the participant sample is drawn from
three different programmes. It is, therefore, a matter of interest to know if the mean
indicators across the different groups are same. Treating the participant sample into
three different groups, we apply ANOV A to test the null hypothesis that the mean
indicator across the different groups, including non-participants, is same. A rejection
oft-test would imply that the sample means are different for different groups. Further,
the tests are also supported by graphical representation of the distribution functions
across participants and non-participants and also across different categories of
population.
5.3.2.2 Difference between Population Groups - Non-parametric Approach
We apply the non-parametric tests to the null hypothesis of equality of distributions
between population groups, pertaining to various indicators. For example, we denote
the population distribution of the level of living index of the members in the
microcredit programme by F(X) and the distribution for the non-members byG(X).
Let the two samples are denoted by X,, X 2 , ••• , X11
andY,, Y2 ••• , Ym respectively.
We begin with commonly used tests for the equality of two distributions. The
Kolmogorov-Smirnov (KS) is used to test the null hypothesis of equal distributions
Chapter 5
against the very general alternative that the distributions are unequal. KS test statistic
measures the maximum difference between the two empirical distribution functions.
Large values of the statistic are evidence against equal distributions and lead to the
rejection of null hypothesis. The exact null probability distribution is available in
tabulated form for small samples. We use critical values based on the asymptotic
distribution, as our sample size of members and nonmembers are both sufficiently
large.
KS test is often used for preliminary studies of data since the alternatives involved are
very general. To test the null hypothesis against the more specific alternative of
stochastic dominance of the non-member distribution, F(X) ~ G(X) for all X ,
F(X) > G(X) for some X, we use the more powerful Wilcoxon-Mann-Whitney
(MW) test. The MW statistic is defined as the number of times an X precedes a Y in
the combined ordered arrangement of the two independent random samples X1
,
X 2 , ••• , X n and Y1 , Y2 ••• , Ym into a single sequence of m + n = N observations,
in ascending order of magnitude. We use critical values based on the asymptotic
distribution of MW under the null hypothesis. This has been shown to be normal with
mean and variance. The null hypothesis is rejected for large values of the statistic. The
test performs particularly well as a test for equality of means (or medians).
5.3.2.3 Examining Targeting by Probability Regression
We employ probit probability regression to estimate the probability of joining
microfinance programme by analysing a number of household and village level
characteristics. Use of probit or logit regression is on the basis of the assumption
regarding the distribution of sample variance. We have a fairly large sample and every
household was free to join the programme. Whether the household joins or not is,
therefore, a case of self selection issue. The latent threshold limits of joining or not
joining the programme are unobserved for any particular household. In this context,
we assume sample variance to be normally distributed and therefore, we employ
probit regression. When & is normal E(& 1 x) = 0 and var(& 1 x) =I , the pdf is
&2 r I 12 ¢(&)= ,1.-exp(--) andcdfis <D(&)= ~exp(-)dt (Long, 1997).
"2" 2 "'-v2rr 2
Chapter 5
Using the probit regression, we estimated the predicted marginal effects and the
predicted probabilities of participation, given X variables. Plotting the predicted
probabilities across per capita income will give us idea about the probability of
participation of households at different points of the income distribution. Therefore,
we can estimate the probability of participation for the households below a threshold
level of living. If the probability of participation below this threshold level is higher
than those above the threshold level, then the programme is well targeted. Therefore,
we can examine the depth of outreach of the programme by using probability
regression.
5.3.2.4 Examining Targeting by "Level of Living" Index
In view of the growing realization that income poverty only partly explains household
vulnerability, researchers have suggested a number of indicators combining economic
as well as social indicators which enhance household capacity to realize a subsistence
level of living. The emphasis on a variety of indicators stems from a realization that
poverty is multidimensional and that any one indicator would not be capable of
capturing poverty levels across different regions and different population groups. A
combination of such socio-economic indicators may help construction of an index that
may be able to explain the relative levels of poverty across households (Henry eta!.,
2000). In the framework of CGAP poverty assessment in microfinance programmes, a
level of living index is constructed using principal component analysis. The index is
constructed using the following variables: type of roof of house, type of wall of house,
type of floor of house, workforce participation rate, any salaried person in the
household, own land, value of productive assets such as livestock, farming assets and
transport assets, annual per capita income, monthly per capita expenditure on clothing
and footwear, education of principal earning member, adult literacy ratio and level of
food security of household during the year. The choice of variables used in the
analysis was based on CGAP proposed indicators 4 and on the basis of field
observation of what captures poverty in this region and on the share of total variance
in the set which was accounted for by the first principal component, which we used as
our index5. To be comprehensive, at least one variable was chosen from each section
4 See Henry et a!., 2000. 5 The level of living index provides a tool to calibrate relative poverty-the extent to which a household is worse off or better off compared to other households. It does not by itself provide
Chapter 5
of the survey to ensure that all aspects of the households' economic condition were
considered.
The construction of "level of living" index uses the method of principal component
analysis (PCA) for obtaining the first principal component to be used as the composite
index. PCA enables construction of a composite index (first principal component) as a
linear combination of standardised indicators with weights obtained from the eigen
vector corresponding to the largest eigen value. It may be pointed out that all the
components are orthogonal, and the first principal component accounts for the largest
amount of variance in the data matrix. The present study uses only the first principal
component as the composite index, as it has the largest possible explanatory power.
Use of more than one principal component would bring in the problem of aggregation
of orthogonal components which should be avoided due to its dubious theoretical
validity.
We also apply the t-tests and the non-parametric tests to "level of living" index score,
which is used as a proxy for household standard of living, in our analysis. We test for
differences in the mean value of this index across participants and non-participants
and also across the four groups. The fairly strong distributional assumptions, under
which these tests are valid, may not be appropriate in the present study since there
may be differences in the distributions of participants and non-participants, or within
different groups among participants, which are reflected, not in their means, but rather
in the tails of these distributions. The wealthier might be ineligible for rural anti
poverty programmes and the poor might be excluded from them for a variety of social
and economic reasons.
Further, we draw the non-parametric distribution functions of the index scores by
applying k-density estimates for the participants and non-participants. For example, if
the k-density distribution of index scores belonging to participants sample shifts to
right of the non-participants it implies that the later group is worse off than the former.
The exercise is conducted across different member groups as well to understand if
there is any difference across participants belonging to different programmes.
information on the absolute level of poverty, the actual level of depravation of the lowest category of households or the level of affluence of the higher group.
Chapter 5
5.4 ESTIMATION RESULTS
5.4.1 Differences in Household Characteristics
Table 5.1 maps average values of different indicators pertaining to participant and
non-participant households. The exercise has been carried on for the aggregative
sample and also for the samples pertaining to the two districts separately. Further, we
also look at the differences in sample means of participants belonging to different
microfinance programmes, as given in Table 5.2. Tests are applied for the indicators
including demographic structure such as household size and number of children of
age upto 6 yrs; economic status captured by number of usual workers in the
household, agricultural landowned and percent of households reporting self
employment in off-farm business; value of productive assets such as livestock,
farming and transportation; consumption expenditure per capita on food, clothing and
footwear and total monthly consumption per capita; and monthly income per capita. It
can be seen from the Table 5.1 that average household size of the participant
households is significantly higher than non-participants in total sample as well as in
district samples separately. The difference among average number of children upto
age of 6 years is not significant either in Betul or in Sehore. The average number of
usual workers is significantly higher for the member households than the non
members in Betul whereas there does not seem to be much difference across the
groups in Sehore. However, the work participation ratio (workers/household size)
turns out to be high among the non-member households than their counterparts, in
each of the districts as also in aggregate sample. One would, therefore, infer that high
work participation ratio tend to be negatively correlated with programme
participation. Nevertheless, the direction of relationship between programme
participation and work participation rate is a proposition to be examined.
One of the most crucial indicators of economic well-being of rural household IS
agricultural land. This does not seem to be different across the population groups. The
test statistics are not significant for any of the categories. However, percent
households reporting self-employment in off-farm activities (excluding livestock) is
much higher for the participant households than their counterpart in Betul. Indeed,
intervention of PRADAN in promoting poultry, mushroom cultivation and sericulture
among the participating households has helped diversifying their economic portfolio,
Chapter 5
whereas undertaking off-farm activities is a distant possibility for the non-participant
households. However, the sample from Sehore indicates that the overall percent of
households reporting self-employment is higher than that of Betul. Understandably,
the locational factors contribute to creation of opportunities for diversifying
production and employment choices and thus, Sehore being situated near a town has
greater opportunities for undertaking off-farm activities. But there does not seem to be
much difference among the participants and non-participants in terms of off-farm
activities undertaken. It was also observed in the study that there was not much extra
effort by the microfinance programmes in promoting off-farm activities among the
participant households in Sehore. The activities of the participants were mostly self
induced which might have been indirectly supported by access to finance or market
exposure training provided by the self-help group promoting institution (SHPI). One
would, however, observe higher incidence of off-farm activities by member
households than the non-member households in both the regions and also at the
aggregate level.
One of the direct gains from participating in micro finance programmes in rural areas
seem to be increase in livestock holding which is significantly different for the
participants than the non-participants. Most of the Bank finances provided to the
participants are for livestock and the household's demand for livestock, particularly
buffaloes, is not only for its net contribution to household earning but also for the
purpose of consumption needs. The difference in value of livestock assets across the
groups is well noticeable in both Betul and Sehore. Value of transportation assets is
also higher for the participant households than the non-participants in all the regions
and also at the aggregate level. But the pattern is not similar in accumulation of
farming assets. Overall, the non-participants have higher values of farming assets than
the participants and the difference is mainly due to the high difference observed in
Sehore.
Indicators on consumption items, particularly clothing and footwear, seem to obtain
high average values for the participant households than the non-participants. Many of
the respondents reported that increase in sociability in the recent years (which may be
attributed to the effect of television, increasing schooling of children, interacting with
officials through participating in group-based activities etc.) has led to higher
Chapter 5
spending on clothing and footwear. Noticeably, in Betul many households have
reported expenditure on footwear, which they were not doing before their
participation in self-help groups (SHGs). Similar is the case for clothing of children,
particularly those who are going to school. Attending meetings of SHGs has
positively influenced many households to invest in such necessities of social life.
However, overall monthly consumption per capita is higher for the non-participants
than those of the participants. Similarly, at the aggregate level, average monthly per
capita income of the non-members is higher than that of the members. In Betul,
average income is significantly higher for the non-participants than the participants.
But, the difference is not statistically significant in Sehore, although the average
income is higher for the non-participants than the participants. Indeed, both the
regions report higher monthly per capita income (MPCINC) for the non-participants
than those of the participants. However, the aggregative sample shows that the
difference rarely becomes significant at only 10 percent level. Table 5 .la reports the
non-parametric tests for the two indicators, MPCE and MPCINC, which may reflect
that the difference in population groups may not lie in average but it may be there in
the tails of these distributions. Both K-S test and M-W test reveal that the difference
of MPCE between the participants and non-participants is not statistically significant.
In contrast to that significant difference is noted in MPCINC for the aggregative
sample and also for Betul. Sehore does not report any significant difference in
MPCINC between the two groups. However, the average MPCINC is higher for the
sample population in Sehore than that ofBetul.
One may look at the empirical distribution functions drawn for the two indicators,
MPCE and MPCINC, to understand their behaviour across population groups and
regions. Figure 5 .I reveals that the two distributions cross at a point which implies
that at the lower end of the distribution, members' average consumption is higher than
that of the non-members. But later on the non-members dominate over the members.
Similar patterns are observed for the two regions as well, as given in Figure 5.1a and
5 .I b. Figure 5 .I c shows the distribution functions across non-members and members
belonging to different programmes. It reveals that the members in SWASHAKTI
programme dominate over all groups upto a certain point, but later non-members
dominate over all the member groups. The relatively poorest are the PRADAN
Chapter 5
members as revealed from the figure that the distribution function is way behind all
other groups. Similar exercise is made for mapping MPCINC. Figure 5.2 shows that
the non-members have clearly an edge over the members. At all points along the
distributions, the non-members' curve lie to the right ofthat ofthe members'. Similar
pattern is observed for the regional sample as well (Figure 5.2a and 5.2b). However,
Figure 5.2c represents a criss-cross of distribution curves across members belonging
to different programmes and those of non-members. The distribution of the
SWASHAKTI members' is at the extreme right in most ofthe points while that of the
PRADAN members' is at the extreme left.
Noticeably, the behaviours of multiple indicators indicate that the distribution of
economic well-being for the two groups differs across different indicators and we
cannot come to any conclusion about targeting from examining these indicators
differently. One way is to look at probability of participation with per capita income,
while controlling for other covariates of household (Diagne and Zeller, 2001 ). The
other way is to construct a level of living index by aggregating the multi-dimensional
indicators and measure the relative poverty between members and non-members
(Henry et al., 2000). Both the methods are attempted in the following sections.
Figure 5.1 Distribution Functions of Monthly Per Capita Consumption Expenditure (MPCE) By Membership -All Households
--tr-- Non-member ------Member
.8
.6
.4
.2
0
0 500 1000 1500 MPCE
'§' 0. E "(:' 0.
Chapter 5
Figure S.la Distribution Functions of Monthly Per Capita Consumption Expenditure (MPCE) By Membership in Betul
--t:s-- Non-member -~-Member
.8
.6
.4
.2
0
0 500 1000 1500 MPCE
Figure 5.1b Distribution Functions of Monthly Per Capita Consumption Expenditure (MPCE) By Membership in Sehore
--t:s-- Non-member -~-Member
.8
.6
.4
.2
0
0 500 1000 MPCE
Chapter 5
Figure S.lc Distribution Functions of Monthly Per Capita Consumption Expenditure (MPCE) By SHG Type -All Households
--- Non-member --e---SGSY --6----- Pradan --><--- Swashakti
.8
.6
.4
.2
0
0 500 1000 1500 MPCE
Figure 5.2 Distribution Functions of Annual Per Capita Income (PCINC) by Membership
--6----- Non-member _...,___ Member
.8
.6
.4
.2
0
0 5 10 15 20 Annual pcinc (Rs. 000)
r-·----, ! 157 J
u c: '(3 .e, a.
Chapter 5
Figure 5.2a Distribution Functions of Annual Per Capita Income (PCINC) by Membership in Betul
--ts--- Non-member --<>--- Member
.8
.6
.4
.2
0
0 5 10 15 20 Annual pcinc (Rs. 000)
Figure 5.2b Distribution Functions of Annual Per Capita Income (PCINC) by Membership in Sehore
--b--- Non-member --<>-- Member
.8
.6
.4
.2
0
0 5 10 15 20 Annual pcinc (Rs. 000)
Chapter 5
Figure 5.2c Distribution Functions of Annual Per Capita Income (PCINC) by SHG Type
u c '(3 a.
15:
.8
.6
.4
.2
0
--- Non-member --6--- Pradan
0
--<>--- SGSY _ _,______ Swashakti
5 10 15 Annual pcinc (Rs. 000)
5.4.2 Determinants of Participation: Pro bit Regression Model
5.4.2.1 Data and Variables
20
The pro bit regression 6 is applied to estimate the probability of participation as a
function of level of per capita consumption ofhousehold (as a proxy for income), off
farm activity, work participation rate, years of schooling of principal earner of
household, adult literacy ratio, productive assets, agricultural land (acres), distance
index7, regional dummy and years of operation of different microfinance programmes
in the villages. The level of monthly per capita consumption is divided into five
quintile groups. For the five quintiles we introduced four dummies in the model,
keeping the dummy of the bottom quintile as the default category. Understandably, if
the programme is appropriately targeted towards the poor then the dummies of the
higher categories should be having negative coefficients, as compared to the default
category.
6 The model assumes normality in distribution of sample variance. 7 See Appendix for details about construction of distance index.
,----, I 159! L_____,
Chapter 5
We have observed from the descriptive statistics that the percentage of households
reporting self employment in non-agricultural activities is higher for the participating
households than the non-participants. However, the variable is introduced as a dummy
to capture the unobserved quality of the household for taking risk in non-farm
activities or the entrepreneurial attitude of any member in the household, which may
induce a person in the household to become a member of the group. Becoming a
member might have helped her/him in getting access to resources for which she/he
could have been able to realize her/his expected earnings. If the expected probability
coefficient is positive and significant then it would confirm that the households with
strong motivation for non-farm activities have higher probability of participation in
the programme.
Work participation ratio (ratio of workers to total household size) captures the stock
of labour in the household. Higher the work participation ratio, lower will be the
dependency ratio and therefore, average per capita earnings will be higher.
Households with higher work participation ratio tend to be in the upper quintile of
income distribution. So the probability coefficient of participation is expected to be
negative with increase in work participation rate in case the programmes are targeted
to the poor.
Education, captured by the indicator years of schooling of principal earning member,
helps enhancing opportunities and economic prospects of the person. Importantly, it
also increases access and participation in different institutions in the village. Literacy
ratio8 (ratio of literate adults to total household members) reflects the capacity of the
household in terms of getting access to the village institutions as well as being aware
about the anti-poverty programmes and policies meant for upliftment of the rural
households. One would, therefore, expect a positive coefficient of probability of
participation for both the educational indicators.
Value of productive assets including farming, livestock and transportation represent
household non-liquid wealth and in the model they are used as control variables.
Owning of agricultural land is another non-liquid wealth which is also used as a
control variable. Distance is a major factor in accessing information and also
8 We have excluded the literacy rate of children (age below 15 years) because the participating decisions in village institutions is taken by the adults.
Chapter 5
accessing village institutions for a rural household. Inaccessibility to outside village
institutions, say local agricultural mandi 9 , local hat 10 or nearest town makes a
household vulnerable to local middleman. More importantly, remoteness restricts
flow of information about different programmes and schemes to the targeted sections.
But capturing distance effect by using a particular distance variable brings biasness in
regression equation as it may overestimate the effect while there are other points
which also influence the household or the households in the entire village. So we
construct a distance index by considering multi-point distances from the village,
measured in kilometers, such as distance of village from block office, nearest town,
bus stop, agricultural mandi and local hat. The hypothesis is that with increase in
score of the distance index household's probability of participation in programme
declines.
Our sample represents two regions and we include a dummy for the region Sehore
which has been selected to represent agriculturally prosperous region and it is also
different because of its better access to state level institutions and better infrastructure
facilities, as compared to Betul. The hypothesis to be examined is whether there exists
any regional effect in participation of programme which will account for the
unobserved effects that the region is representing for.
Finally, we have included three variables for three programme effects which are
introduced as number of years of operation of the programme in the village. For
example, a positive and significant coefficient will imply that with increase in
operation of the programme the rate of participation in the village will increase.
5.4.2.2 Regression Results
The regression coefficients and predicted marginal probability of participation in
group-based credit programmes is presented in Table 5.4. The coefficient of dummy
for the 2nd quintile is positive but insignificant. The coefficients of dummies for the
3rd and 4th quintiles are positive and highly significant. Importantly, the values of the
coefficients are increasing, the figures being 0.476 and 0.571. One would, therefore,
see increase in the marginal effects of probability of participation with increase in
9 A1andi is a government administered trading centre. 10 Hat is the local weekly or monthly regular market place.
Chapter 5
level of living upto a certain point, i.e., fourth quintile of population distribution and
then it declines. Noticeably, the marginal effect of the dummy for the fifth quintile is
negative, although not significant.
A household's probability of participation increases if any person from the household
is having entrepreneurial attitude, which is confirmed from the positive direction of
the indicator on self employment in off-farm business. The coefficient is also
statistically significant at 5 percent level. One might say that this is post facto
situation. Indeed, we do not have baseline data in which we have information about
the participant; whether he/she was interested to join the group for undertaking off
farm business. It is generally observed that the households with some inclination to
start business or other non-farm enterprises are incapable of doing so because of
restricted knowledge of product or factor markets, procedures of starting business or,
most importantly, shortage of capital. Propaganda of microcredit programmes
motivated these households to join the programme and then they took the initial step
in starting off-farm business by borrowing from the group. For example, one
participant reported that she wanted to support her son to start a business (selling
items by moving from place to place in a bicycle), but she had no capital. At that time
Swashakti SHGs were formed in the village and she joined a group with the hope that
she can borrow from the group at a later date. Indeed, after some time she borrowed
from the group for the purpose of business support to her son, and now their earnings
from business has grown significantly. Nevertheless, many households reported that
main purpose of joining the group was to avail credit for income generation purposes,
besides fulfilling consumption needs. So the dummy for self employment activities
represents that with motivations for non-farm work the probability of participation
increases.
Work participation is inversely related with probability of programme participation,
the coefficient being significant at I 0 percent level. This indicates that higher the ratio
of workers in the household lower is the probability of participation in programme.
As work participation rate is higher among the better off households, it implies that
participation rate of such households in the programme is low.
The coefficient of education of earning member turns out to be negative and
significant which implies that higher the level of education lower will be the incentive
Chapter 5
to join the programme. Although this contradicts our hypothesis of positive influence
of education on programme participation, it may be explained by the fact that
education increases economic opportunities of a person and if the returns are higher in
alternative occupation then he/she will not prefer to join the programme, because the
opportunity costs of joining the programme may be higher for him/her. In our context,
higher level of education has been observed for the persons with higher level of
household income or wealth and the participation rate of these households is low as
compared to others. On the other hand, educational attainments are very low among
the poor or the population belonging to the lower strata of income distribution. With
the population highly skewed in terms of educational attainment (high concentration
of persons with higher schooling years in higher income strata) the probability of
participation, therefore, declines with increase in level of educational attainment.
However, this may not be seen as a consequence of higher schooling years, rather an
interaction effect of higher schooling years and higher income. It may be noted that
the effect of literacy ratio on probability of programme participation is positive
although it is not statistically significant.
Distance index has negative effects on probability of participation in the programme.
For instance, if the index has higher values for a particular village then the households
in that village have lower probability of participation. This is primarily due to the fact
that with increase in distance in accessing the village, the cost of communication for
the programme agents increases and the level of intervention decreases. So
penetration of the programme is restricted to few households in the village.
The coefficient of the dummy for district Sehore is negative and significant at 10
percent level. We found that penetration rate is higher in Betul than Sehore, although
the later has geographical advantages over the former. However, the region effect
should not be seen in isolation. We must take into account the interaction of
programme effect with the region effect. One would notice that the higher penetration
in Betul villages is due to greater amounts of efforts of PRADAN, whose processes of
SHG formation and management is much more rigorous and professional than the
agents of programme promotion in Sehore. Another factor is that the density of
households in Betul is much less than Sehore. However, probability of participation
differs across the regions, which can mostly be attributed to the nature of institutions
Chapter 5
supporting the programme. Longevity of operation of programme in the village has a
positive influence on probability of participation in the programme, particularly for
the households in the villages where the PRADAN programmes are in operation.
Unfortunately, the same effect is not seen in case of the SGSY programmes or the
SW ASHAKTI programmes.
The predicted probabilities, estimated from the above regression model, are shown in
Figure 5.3 across consumption quintile groups, for the regions of Sehore and Betul
separately. Further, we also estimated probabilities across annual monthly per capita
income which is plotted in Figure 5.4. Using either consumption or income as an
indicator of economic well-being one would observe that the predicted probabilities
are low for the households belonging to the lower section of population distribution.
The probability of participation increases with increase in economic well-being upto
the fourth quintile in both Betul and Sehore. Of course, the probabilities decline at the
top most quintile but that may not be a concern since many households in the upper
section will never participate in the programme and also they are discouraged to
participate in the programme by the programme agents.
Figure 5.3 Probability of Participation of Household in SHG Programme by Consumption Expenditure Quintiles
0.90
0.80
c ~ 0.70 .9-~ 0.60-
"' a. 0 0.50
i 0.40
"' .0 £ 0.30
0.20
0.10 .
------------··---------- Sehore -Betul
0.00 L------~-------------------~ 2 3 4 5
MPCE Qunitile (District wise)
----- ---·- --- ------- ------------------------- -- --
Chapter 5
Figure 5.4 Probability ofParticipatiou ofHousehold iu SHG Programme by Annual Per Capita Income (Rs. 000)
.8
E' .6 Q)
E Ol
..r:. (/)
'c-a.. .4
.2
0
0
0 0
oO
2.5
0 0
0
0 0 0
0
0
5
0
0
0
0
0
0
0 0
0 0
00
7.5
oo
0
0 0
0 0 0
0
0 0 0
0 0 0 0
0 0
0 0
0
10 12.5 15 17.5 pcinc1
In Figure 5.4 we have added three lines, representing the minimum expenditures
required for different levels of living for the rural population, along with the predicted
probabilities. These three vertical lines represent national level estimates from the
latest NSS quinquennial round of consumption expenditure, pertaining to the year
2004-05 (NSS, 2006). This is also the survey year of this research and therefore, the
estimates do not require adjustment for prices. The first line represents Rs. 3240 per
capita per annum, which is calculated as a benchmark for extreme low levels of I iving
(Rs. 9 per day); the second line is the bench mark for subsistence level ofliving ofRs.
4380 per capita per annum (Rs. 12 per day); and the third line represents Rs. 8040 per
capita per annum (which belongs to the 801h percentile of the all-India distribution of
per capita consumption of the rural population) 11• NSS uses consumption as proxy for
income. In our context, we have the income figures and therefore, the lines are drawn
on incomes rather than consumption 12•
11 The NSS consumption expenditure figures are given as per capita per month, such as Rs. 270, Rs. 365 and Rs. 690 respectively. 12 See appendix for discussion on reliability and consistency of household income estimation and its correlates with other household indicators.
Chapter 5
Noticeably, much of the sample concentrates below Rs. 12 per day. However, it is a
matter of concern that probability of participation is very low below this level of
income and the probability line is rising with rise in income. One would also see that
below the income level of Rs. 9 per day there are many households whose probability
of participation in near to zero. These are the section of population who are very poor
and they are excluded from the programme. So, this raises a question on use of
microfinance programmes as a substitute for poverty alleviation programmes.
However, many would not accept this result because of the use of income as the only
indicator of economic well-being. So the following section uses an index, constructed
by combining multi-dimensional indicators, for measurement of targeting.
5.4.3 Level of Living Index
The level of living index is constructed using principal component analysis. We use a
wide number of indicators pertaining to different aspects of household level of living
and the final set of indicators are retained, for the first component, on the basis of the
significance of the factor loadings and the eigen values. This has been discussed in
section 5.3.2.4. The variance of the first component, pertaining to the largest eigen
value, accounts for 34 per cent of the total variation in these variables. The factor
loadings pertaining to the first component are given in Table 5.6 and the variance of
each component and associated eigen values are in Table 5 .5. We use the first
principal component as our index of level of living13. In fact, the factor loadings are
weightages assigned to individual indicators which reflect the correlation between the
indicator and the composite index. Higher positive correlations imply better
appropriateness of the indicators. It may be seen from the table that all the indicators
are having factor loadings greater than 0.3 (which is showing the high correlation of
the indicator with the component) except for the indicator on work participation ratio.
However, the indicator is retained in the analysis because of the importance of the
indicator for enhancing household earnings, although it is not statistically significant.
13 We find that estimated distribution functions for the annual household expenditure on clothing and footwear behave very much like the index as a whole. The index also behaves very much like the per capita income. Both are highly correlated. This is consistent with other studies that have used this methodology to measure standards ofliving (Henry et at., 2000).
[i66]
Chapter 5
The distribution of level of living index for members and non-members were
compared using a variety of parametric and non-parametric statistical tests. Mean
values of the index by region and by group membership is given in Table 5. 7. It also
presents the results of t-test of difference of group means of the index. The null
hypothesis of equal means cannot be rejected for the entire sample as a whole, while
we find the difference to be statistically significant at .06 level in one region (Betul).
It is worthwhile to examine whether the microcredit programme targets households in
the middle of the distribution and excludes both tails. Then we need to estimate the
entire population distribution for members and non-members in order to identify the
nature of targeting in the programme. Figure 5.5 represents the distribution functions
of the index for members and non-members. Noticeably, the two functions cross each
other at a point near to the mean value of the index, with the curve of the members
lying to the right of the non-members, and after the crossing the distribution of non
members shift to the right of the members. This implies that the bottom sections of
population distribution are not well targeted in the programmes. This is also
confirmed from the region-wise distribution functions of the index as shown in Figure
5.5a and Figure 5.5b.
We conducted the non-parametric tests by applying K-S test and also M-W test, the p
values are shown in Table 5.8. Noticeably, the tests reject the null hypothesis of equal
distribution of members and non-members for the whole sample. Both the tests are
statistically significant at .05 level. Across the regions, we obtained similar results in
Betul, but the hypothesis of equality of distributions of members and non-members
cannot be rejected in Sehore.
We also examined the difference in means of the index and also the difference in
distribution functions of the index across programmes by applying ANOV A as well
as non-parametric tests respectively. Table 5.9 gives the results of both the tests. One
would simply derive the conclusion that the null hypothesis of equality of
distributions of the level of living index is rejected, as both ANOVA and K-W tests
are significant at .05 level. This is also reflected in the distribution functions, as
shown in Figure 5.5c.
.8
Q) .6 0
()
"' X Q) "0 c ~ .4 a.
.2
0
.8
~ .6 0 ()
"' X Q) "0 c
;::.. .4 0..
.2
0
Figure 5.5 Distribution Functions of Level of Living Index by Membership Status- All
--f!s--- Non-member --<>---Member
-1.000 0.000 1.000 2.000 3.000 4.000 index score
Figure 5.5a Distribution Functions of Level of Living Index by Group Membership- Betul
--f!s--- Non-member _....,___Member
-1.000 0.000 1.000 2.000 3.000 4.000 index score
Chapter 5
5.000
5.000
.8
Q) .6 ....
0 0 (/)
X Q) "0 r:: ~ .4 a.
.2
0
.8
Q) .6 ....
0 0 (/)
X Q) "0 r:: '=- .4 c.
.2
0
Figure S.Sb Distribution Functions of Level of Living Index by Group Membership- Sehore
--es-- Non-member --+-- Member
-1.000 0.000 1.000 2.000 3.000 4.000 index score
Figure S.Sc Distribution Functions of Level of Living by Group Membership- All
--- Non-member ----SGSY --es-- Pradan --><-- Swashakti
-1.000 0.000 1.000 2.000 3.000 4.000 index score
Chapter 5
5.000
5.000
Chapter 5
Estimated kernel densities for the level of living index can be used to get a better idea
of the set of households that are relatively neglected by the programme, and those that
gain the most from it. Estimated densities for the whole sample as well as for regions
are shown in Figures 5.6 to 5.7. For the whole sample, when both densities are rising
the crossing point is less than zero while when both densities are falling the crossing
point is at the tails (near 2). Therefore, the households between these two crossing
points are those selected by the programme14. One can also see the crossing points of
the densities across the different programmes and that of the non-members (Figure
5.8).
~ 'iii c: <I> a
0
-2
Figure 5.6 Kernel Density Estimates Level of Living Index by Membership Status - All
/\ I \ I \
\ \ \ \ \
0
\ \ \ \ \ \ \ \
' ' \ \
\.
' ~ :,..
2 Index score
---4
--- Non-member ----- Member I
14 Similar results were obtained by Somanathan and Dewan (2003).
6
Chapter 5
Figure 5.7 Kernel Density Estimates by Membership Status- District-wise
>:"!:V (/) . c: Q)
0
Betul Sehore
,., ' \. " ' \
\ \ \
' \
-2 0 2 4 -2 0 2 4
1-- Non-member ---- Member I 1-- Non-member ---- Member I Index score
Graphs by district
z. ·u; C:...r Q) •
0
0
-2
Figure 5.8 Kernel Density Estimates Level of Living Index by Group Membership- All
I \
0 2 Index score
4
--- Non-member ----- SGSY
ere
-···- Pradan Swashakti
6
Chapter 5
Henry et a/. (2000) suggested measurement of targeting by looking at the relative
poverty of members and non-members, from the index score. First, quartiles of the
index score are calculated for the non-member households. Second, the cut-off scores
of the quartiles for the non-members are applied as cut-off scores for grouping
member households. Ideally with no difference in distribution among member and
non-member households, the quartile groups will be exactly equal. Evidently, the ratio
of members to non-members in the lowest quartile turns out to be 0.13/0.25, while the
ratio is greater than I in 2"d and 3rd quartiles (Figure 5.9). One would, therefore,
conclude that the poorest households are excluded from the programme, as the ratio of
non-members to members is less than I in the bottom quintiles, while it is greater than
one in upper quintiles.
Figure 5.9 Percent of Participant Households Corresponding to Quartiles of NonParticipant Households Classified by Level of Living Index
35 o Non-member
30
25 0. ::)
e (.!)
20 <: 0
N ::) 0.
15 0 Cl.
0 ~ 0
10
5
0 01 Q2 03 04
Poverty Group
Note: First, quartiles are calculated for the non-member households using the household level of living index scores. Second, the cut-off scores of the quartiles for the non-members are applied as cut-off scores for grouping member households. Ideally with no difference in distribution among member and non-member households, the quartile groups will be exactly equal.
Chapter 5
5.5 CONCLUSION
Our results indicate that probability of participation is low at lower end of income
distribution and it increases with increase in per capita income of the household. Of
course, it declines with very high level of per capita income of household. But these
households may as well never participate in the programme. Since microfinance
programmes aim at meeting the credit needs of low income clientele, or supporting
micro enterprises, they may not be attractive for the high income clients. They will
only participate if the returns from joining the programme will be higher than their
present occupation. Some rich households do not join because they are discouraged by
the programme agents to join the programme. Besides, it is below the status, in the
village, for the women from rich households to join the group and save a small
amount.
Probability of participation is high in Betul than Sehore. But the region effect cannot
be explained in isolation. The interaction of programme characteristics has to be
examined before drawing any conclusion about the region effect. We found
programme effect (PRADAN) to be a significant determinant in increasing probability
of participation. In other words, the design and process variables of the programme do
determine the household participation in the group. However, we have no conclusive
evidence on whether peer selection affects the exclusion of very poor from the
programme.
Chapter 5
Table 5.1 Mean of Selected Indicators by Membership Status of Household
t-test Nonmembers Members (p-value)
Total Sample
Household size 5.3 6.5 0.000
No. of children of age upto 6 yrs. 1.3 1.3
No. usual workers 3.0 3.4 0.053
Own land (acre) 2.2 2.4
Value of livestock (Rs. 000) 4.3 7.2 0.036
Value of farming assets (Rs. 000) 10.9 8.7
Value of transport vehicles (Rs. 000) 1.4 3.5 0.065
Monthly expenditure on food per capita (Rs.) 178.0 177.6 Monthly expenditure on clothing and footwear per capita (Rs.) 13.7 16.9 0.063
Monthly consumption expenditure per capita (Rs.) 319.2 298.1
Monthly income per capita (Rs.) 414.6 356.3 0.092
%households self-employed in off-farm business 11.4 19.5
Betu1
Household size 4.4 6.2 0.000
No. of children of age up to 6 yrs. 1.1 1.4
No. usual workers 2.6 3.4 0.008
Own land (acre) 2.1 2.6
Value of livestock (Rs. 000) 3.2 5.2 0.007
Value of farming assets (Rs. 000) 0.8 1.9
Value of transport vehicles (Rs. 000) 0.5 1.0
Monthly expenditure on food per capita (Rs.) 149.5 137.8 Monthly expenditure on clothing and footwear per capita (Rs.) 8.9 12.0 0.024
Monthly consumption expenditure per capita (Rs.) 243.6 212.3
Monthly income per capita (Rs.) 350.4 276.6 0.014
%households self-employed in off-farm business 2.6 18.2
Sehore
Household size 5.9 6.7 0.083
No. of children of age upto 6 yrs. 1.4 1.2
No. usual workers 3.2 3.4
Own land (acre) 2.2 2.3
Value of livestock (Rs. 000) 5.1 8.8
Value offarming assets (Rs. 000) 18.6 14.1
Value of transport vehicles (Rs. 000) 2.2 5.4 0.079
Monthly expenditure on food per capita (Rs.) 199.7 209.3 Monthly expenditure on clothing and footwear per capita (Rs.) 17.4 20.7
Monthly consumption expenditure per capita (Rs.) 376.7 366.2
Monthly income per capita (Rs.) 463.3 419.4
%households self-employed in off-farm business 18.0 20.6 Source: Computed from Survey data
~-----....,
Ll2iJ
Chapter 5
Table 5.1a Non-Parametric Tests of Equality of Distributions of MPCE and PCINC H h ld b M b h" St t across ouse o s >Y em ers 1p a us
Kolmogorov-Smirnov Mann-Whitney Re11:ion (p-value) (p-value)
MPCE Betul 0.201 0.241
Sehore 0.113 0.270
Total 0.240 0.688
PC INC Betul 0.048 0.014
Sehore 0.640 0.524
Total 0.049 0.053 Source: Same as Table 5.1
T bl 52 S I a e e ecte d I d" b T n tcators >Y Lypeo fSHGM b h" fH em ers tp o h ld ouse o
Members Non- An ova
member SGSY PRADAN SWASHAKTI (p-value)
Average Household size 5.3 6.5 6.3 6.7 0.001 Average No. of children of age upto 6 yrs. 1.3 1.1 1.4 1.3
Average No. usual workers 3.0 3.4 3.5 3.3 Average Own land (acre) 2.2 2.2 2.6 2.5
Average Value of livestock (Rs. 000) 4.3 11.3 4.6 5.5 0.000 Average Value offarming assets (Rs. 000) 10.9 16.2 1.8 7.8 Average Value of transport vehicles (Rs. 000) 1.4 4.4 1.1 4.8 0.016 Average Monthly expenditure on food per capita (Rs.) 178.0 184.3 144.8 203.2 0.006 Average Monthly expenditure on clothing and footwear per capita 13.7 18.4 11.6 20.5 0.000 Average Monthly consumption expenditure per capita (Rs.) 319.2 312.5 218.1 362.0 0.000 Average Monthly income per capita (Rs.) 414.6 359.2 280.0 428.2 0.007 %households food insecured in past 12 months 48.9 38.3 35.0 35.0
% head of households ill iterate 57.8 56.7 45.0 48.3 %head of households >=primary education 32.2 31.7 28.3 36.7 %head of households >=secondary education 5.6 6.7 1.7 8.3 % households with any adult >=primary education 55.6 65.0 61.7 71.7 % households with any adult>=secondary education 11.1 16.7 10.0 21.7
%households with Katcha roof 86.7 81.7 85.0 86.7
%households with Katcha walls 73.3 61.7 80.0 70.0
%households with Katcha floor 88.9 91.7 90.0 91.7
% ST households 44.4 33.3 96.7 0.0
% SC households 12.2 15.0 1.7 38.3 %households with self employment in off-farm business 11.1 18.3 18.3 23.3 Source: Same as Table 5.1
Chapter 5
Table 5.3 Share of Sources oflncome by District and Membership Status
Betul Sehore Sources of Income Non- Non-(Net) member Member Total member Member Total Agriculture 14.6 14.1 14.4 18.8 18.6 18.8 Home produce 1.8 3.7 2.7 2.8 2.2 2.6 Agriculture labour 21.3 27.9 24.5 14.2 7.2 12.3 Non-agriculture labour 27.8 24.6 26.3 20.8 27.3 22.6 Salaried income 5.4 3.5 4.5 15.3 15.9 15.4 Livestock 9.6 3.8 6.8 12.0 15.0 12.8
Other Off-farm business 1.3 9.2 5.0 11.1 7.9 10.2 Other income 18.2 13.3 15.9 5.1 5.9 5.3 Total 100 100 100 100 100 100 Source: Same as Table 5.1
Table 5.4 Pro bit Regression: Determinants of Participation and Marginal Effects
Dependent Variable: Participation of Household in SHG Programme (l=Yes, O=No)
Coefficient estimates Robust
Mean Std. Err. P>z HH in mpce quintile 2nd (I ,0) 0.256 0.284 0.369 HH in mpce quintile 3rd (1,0) 0.476 0.282 0.091 HH in mpce quintile 4th (1,0) 0.571 0.291 0.050 HH in top mpce quintile (I ,0) -0.265 0.288 0.358 Self employment in off-farm business (1,0) 0.614 0.276 0.026 Ratio of workers to hhsize (Ratio) -0.814 0.468 0.082 Log of years of schooling of principal earner -0.314 0.150 0.037 Ratio ofliterate adults to hhsize (Ratio) 0.638 0.425 0.134 Value of farming assets (Rs. 000) -0.002 0.003 0.504 Value oflivestock assets (Rs. 000) 0.037 0.011 0.001 Value of transport assets (Rs.OOO) 0.042 0.019 0.025 Own land (acres) -0.023 0.032 0.473 Years ofSGSY operation in the village -0.016 0.062 0.801 Years ofPRADAN operation in the village 0.161 0.069 0.019 Years ofSWASHAKTI operation in the village 0.092 0.082 0.263 Distance index -0.570 0.316 0.071 Dummy for district Sehore -0.695 0.416 0.095 Constant -0.005 0.629 0.993
Log pseudolikelihood -131.73 N Wald chi2 Prob > chi2 McFadden's R2
Cragg & Uhler's R2 (*)Discrete change of dummy vanable from 0 to I Source: Same as Table 5.1
[i76'l
262
66.59 0.000
0.19
0.29
Predicted marginal
effect
dF/dx
0.089 0.170 0.207
-0.085 0.227
-0.273 -0.105 0.213
-0.001 0.012 0.014
-0.008 -0.005 0.054 0.031
-0.191 -0.242
Chapter 5
T bl 55 E. a e 12en VI a ues an dV ar1ances o fP. rlnCipa IC t omponen s Cumulative variance
Components Eig_en value explained I 4.801 0.343
2 1.559 0.454
3 1.327 0.549 4 1.130 0.630
5 1.018 0.703 Source: Same as Table 5.1
T bl 56 F a e . actor L d. ~ th F. t P . oa m2s or e IrS r1nc1pa IC omponen t Indicators Factor loadings Roof type 0.453 Wall type 0.588 Floor type 0.636 Ratio of workers to household size 0.223 Any person salaried in HH (I ,0) 0.408 Own land (acres) 0.605 Value of livestock assets (Rs. 000) 0.506 Value of transport assets (Rs.OOO) 0.663 Value of farming assets (Rs. 000) 0.633 Annual per capita income (Rs. 000) 0.738 Monthly expenditure on clothing and footwear per capita (Rs) 0.763 Education of principal earning member of household 0.611 Ratio of literate adults to household size 0.627 Level of food security during last 12 months 0.525 Source: Same as Table 5.1
Table 5.7 District-wise Mean Values of Level of Living Index for Members and Non-Members Region Non-members Members t-test (p-value)
Betul -0.493 -0.232 0.067 Sehore 0.210 0.221 0.530 Total 0.001 -0.002 0.253 Source: Same as Table 5.1
Table 5.8 Nonparametric Tests of Equality of Distributions of Level of Living 1 d H h ld b M b h. St t n ex across ouse o s >y em ers 1p a us Reeion KS (p-value) MWJp-valu~
Betul 0.026 0.104
Sehore 0.209 0.306 Total 0.027 0.042 Source: Same as Table 5.1
Table 5.9 Mean Values of Level oj_Livinf( Index B, SHG type A nova Kruskal-Wallis
SHG Type Mean (p-value) (p-value) Non-member 0.001 SGSY 0.128
0.044 0.036 PRADAN -0.222 SWASHAKTI 0.155 All members (SGSY, PRADAN, SWASHAKTI) -0.002 Source: Same as Table 5.1
Chapter 5
APPENDIX
Calculation ofDistance Index
The steps involved in the construction of distance index for the five indicators are appended below: Step 1: Making the indicators scale free
i. Computation of the mean for each indicator. For the /''indicator, the
mean x; would be LX ij IN where X ij is the value of the i'" village for
j'" indicator and i = 1,2, .... .N. N is the number of villages, which is 18 in the present case for each indicator.
ii. Division of each observation of the indicator (X ij) by respective mean
ex;). We thus obtainXij I x;. These may be called the scale free values
of the indicators. Step 2: Adding up the scale free values of the j'" indicators for each village.
D, = I(xijlx;), J=I,2, ..... 5.
Finally, the distance index is D* = D jk, where k is the total number of indicators, .I .I
which is 5, as noted above.
The index shows that higher the value for a village the higher the village will be in a disadvantageous position. The lower the values the more connected the village to the important destinations. Distance of Village from Important Destinations (Km.) and Distance Index
Indicators of Important Destinations
Nearest Local Distance No. Village Block Town Bus Stop Mandi Haat Index I Pola Patthar 14 3.0 0.1 38 3 0.7 2 Mansinghpura 18 7.0 4.0 42 3 1.2 3 Chirmatekri II 6.0 3.0 41 6 1.1 4 Tetar 27.0 16 1.0 62 16 1.9 5 Kuppa 25.0 14 0.1 60 14 1.7 6 Handipani 20.0 8.0 0.1 55 8 1.2 7 Sonadeh 18.0 7.0 0.1 60 7 1.2 8 Phooti Baudi 3.0 6.0 0.1 3 4 0.3 9 Bijora 6.0 6.0 6.0 6 6 0.9 10 Kaudia Chittu 7.0 8.0 8.0 7 8 1.2 II Semli 6.0 5.0 3.0 6 10 0.8 12 Raj_ukhedi 10.0 12.0 5.0 12 12 1.3 13 Manpura 9.0 11.0 1.0 9 11 0.9 14 Rola 12.0 13.0 3.0 12 13 1.2 15 Jahangirpura 5.0 4.0 4.0 5 4 0.6 16 Sekdakhedi 4.0 3.0 2.0 4 3 0.4 17 Takipura 6.0 5.0 5.0 6 5 0.8 18 Konajhir 8.0 7.0 0.1 8 7 0.6
Mean 11.6 7.8 2.5 24.2 7.8 1.0
c.v. 62.3 48.8 95.5 95.2 52.1 40.4