Table 1: List of variables considered in the limited dependent variables models
Name of the variable/notation
|
Description/rationale for inclusion in the models
|
Measurement
|
Choice of the primary cooking fuel (dependent variable Y)
|
Indicates for every family whether firewood is chosen or a basket of clean fuel is chosen as a primary cooking fuel source
|
This is measured as a binary dummy with the usage of firewood getting a value and a different value otherwise.
|
Monthly per capita expenditure (mpce)
|
Indicates how much each member of the family spends every month on a certain set of consumable items.
|
This is measured as a nominal value in INR
|
Monthly per capita expenditure for marginal rural population (mpce_mrp)
|
Monthly per capita expenditure of marginal rural population in terms of energy consumption. The set consists of people whose energy consumption per day is 629 kcal or below for cooking.
|
This is measured as a nominal value in INR
|
Type of district (district)
|
Indicates the nature of the district in terms of availability of banking, road, transportation, health, school, and basic sanitation infrastructure
|
This is measured as a cumulative variable. The value of the dummy for the same district changes over time, i.e., the dummy is time-variant. For instance, the dummy variables are displayed "0" for each month from the beginning of the period up to a certain period (say from the April month of 2000 to April of 2004) and then "1" for each month from the next cumulative period (from May 2005 to May 2009). Measured as a binary dummy where a specific value is assumed if there is an availability of the mentioned infrastructure; otherwise, a different value is considered.
|
Availability of internet (internet)
|
Indicates the availability of information through the internet using a mobile phone, computer, or any other medium. This has been chosen as a proxy for access to sources of information. Information is now mainly accessed through the internet even though radio and television might also be available for the same rural family.
|
The variable is measured as a binary dummy where if there is internet availability, a specific value is given; otherwise, a different value is imposed. Even though often the internet is not a binary dummy because of the degree and quality of the internet connection, in a remote developing country context of India, the families do not have the luxury, nor the physical infrastructure and maintenance service is high in a remote village of India to create a step dummy to the variable internet. In Indian rural families, it is a binary choice, and it will become a step dummy very soon.
|
Nature of liquidity (regsal)
|
Indicates regularity of income or liquidity and proxied by the frequency of expenditure made by families on various items in a month
|
Measured as the number of times a family makes a substantial purchase of family items in a month.
|
Family size (hh_size)
|
Size of a family
|
Measured as a nominal value
|
Characteristics of families (hh_type)
|
Indicates the classification of a family as per NSSO. The villages are differentiated by agroeconomic characteristics in terms of (i) area and population of the village, (ii) cultivated area, types of irrigation, main crop, etc., (iii) small-scale industries, handicrafts, trade, and business, (iv) sources of drinking water, availability of electricity, (v) distances of school, college, health center, hospital, post office, bus-stop, railway station, market, bank, etc., and (vi) facilities for transport and communication, and so on
|
Measured as a scale dummy where the village gets a different range of dummy values of 1 to 3 (low, medium, and high) based on the agroeconomic characteristics
|
Fuel quality/calorific value of fuel for cooking for a family (cv)
|
Indicates the efficiency of the cooking fuel in terms of the energy content. Higher the calorific value, the greater the fuel efficiency
|
This is a nominal value. Fuel consumption is measured in terms of kg for every family. Every fuel has a standard calorific value per kg of fuel. The calorific value is multiplied by the total fuel consumed (in Kg) by the family to get the total calorific value of the family.
|
Presence of social groups (stat reg)
|
Indicates the number of social groups. The larger the number of social groups present, the more diversity and complexity of linkages/dynamics between the groups influences the probability of switching to a particular fuel by a rural family.
|
This is a nominal value indicating the number of social groups in an observation (around the area of the families)
|
Belonging to a particular community (rel)
|
Defines whether the family group belongs to scheduled caste, tribes, or other groups as per NSSO classification
|
Measured as a binary dummy variable where if a family belongs to a scheduled caste or tribe or other groups, they get a specific value and a different value, otherwise.
|
Presence of social/cultural grouping or institutional network (socgrp)
|
Indicates the presence of institutional networks or social groupings like SHGs.
|
Measured as a binary dummy. If there is the presence of a social group, a dummy value is considered; otherwise, an alternate value is considered.
|
** A Note On The Nature of Endogeneity of the Variables – The model takes the chance of the endogeneity into account, which can arise from the variable specifications. Through the set of controls from a large set of pooled regression, gradually regressors are introduced with controls on specific variables to eradicate the chance of endogeneity impacting the model estimates and specifications. After that, likelihood estimators are used, indicating that the model-specific errors are not above or overshooting the standard errors. Through the likelihood estimators, the model-specific errors are minimized to tackle any unknown endogeneity which might still exist in the model even after the imposition of the controls. This error can happen due to the variable observations emerging from the samples of NSSO, which is based on population samples leading to unknown errors and endogeneity. Even that is being reduced through the likelihood estimators.
** A Note On The Nature of Endogeneity of the Variables – The model has taken care of the chance of the endogeneity which can arise from the variable specifications. For instance, wherever there is a chance of a rise in endogeneity, controls are imposed on some of the predictors/regressors so that the endogeneity is eliminated. Through the set of controls from a large set of pooled regression, gradually regressors are introduced with controls on certain variables to eradicate the chance of endogeneity impacting the model estimates and specifications. After that, likelihood estimators are used which indicates that the model specific errors are not above or overshooting the standard errors. Through the likelihood estimators, the model specific errors are minimized to tackle any unknown endogeneity which might still exist in the model even after the imposition of the controls. This existence of the error can happen due to the variable observations emerging from the samples of NSSO which is based on population samples leading to unknown errors and endogeneity. Even that is being reduced through the likelihood estimators.
Table 1a: Logit Regression: 1
|
Coeff
|
Standard Error
|
Z
|
p>z
|
95% Confidence Interval
|
hh_size
|
-0.881
|
0.016
|
-55.51
|
0.000
|
-0.913
|
-0.850
|
hh_type
|
-0.050
|
0.035
|
-1.42
|
0.154
|
-0.119
|
0.019
|
rel
|
-0.159
|
0.057
|
-2.81
|
0.005
|
-0.270
|
-0.048
|
socgrp
|
-0.000
|
0.058
|
0.01
|
0.994
|
-0.112
|
0.113
|
mpce_mrp
|
.000
|
7.53e-07
|
20.90
|
0.000
|
0.000
|
0.000
|
constant
|
-1.639
|
1.860
|
-8.81
|
0.000
|
-2.004
|
-1.275
|
hh_size
|
1
|
(offset)
|
|
group variable considered- hh_type
offset variable (considering fixed effect) – hh_size
level- 1 (group variable- hh_type)
Outcome Table 1
No. of Observation-5000
No. of groups- 4
Integration Points- 7
Log likelihood - -2635.0129
Wald Chi2- 4978.09
Prob> Chi2- 0.0000
Random-effects Parameters
|
Estimate
|
Standard Error
|
95% Confidence Interval
|
rel*
|
0.095
|
0.040
|
0.041
|
0.219
|
socgrp*
|
0.106
|
0.045
|
0.056
|
0.243
|
LR test vs. logistic regression: chibar2(2) = 83.01 Prob>=chibar2 = 0.0000
*specific dummies
** hh_size, hh_type, mpce_mrp, stat reg, internet are controls
Table 1(b): Logit Regression: 2
|
Coeff
|
Standard Error
|
Z
|
p>z
|
95% Confidence Interval
|
hh_size
|
0.114
|
0.016
|
7.36
|
0.000
|
0.084
|
0.145
|
hh_type
|
-0.071
|
0.015
|
-4.51
|
0.00
|
-0.102
|
-0.040
|
rel*
|
-0.091
|
0.052
|
-1.75
|
0.080
|
-0.194
|
0.011
|
socgrp
|
0.010
|
0.017
|
0.60
|
0.548
|
-0.023
|
0.043
|
mpce_mrp
|
0.000
|
7.57e-07
|
22.64
|
0.000
|
0.000
|
0.000
|
constant
|
-1.912
|
0.280
|
-6.83
|
0.000
|
-2.460
|
-1.363
|
group variable considered- rel
offset variable (considering fixed effect) – none, since rel is a group variable, effect considered is random
level- 1 (group variable- rel)
*specific dummy
*stat_reg, internet are the controls
Outcome Table 2
No. of Observation-5000
No. of Groups - 4
Integration Points- 7
Log likelihood - -2668.9793
Wald Chi2- 532.74
Prob> Chi2- 0.0000
Random-effects Parameters
|
Estimate
|
Standard Error
|
95% Confidence Interval
|
rel*
|
0.281
|
0.117
|
0.124
|
0.638
|
LR test vs. logistic regression: chibar2(01) = 15.16 Prob>=chibar2 = 0.0000
*specific dummy
* mpce, mpce_mrp, district, internet, hh size, hh type, stat-reg are controls
Table 1(c): Logit Regression: 3
|
Coeff
|
Standard Error
|
Z
|
p>z
|
95% Confidence Interval
|
hh_size
|
0.133
|
0.016
|
8.53
|
0.000
|
0.102
|
0.164
|
hh_type
|
-0.071
|
0.016
|
-4.47
|
0.00
|
-0.101
|
-0.039
|
rel*
|
-0.086
|
0.026
|
-3.35
|
0.001
|
-0.137
|
-0.035
|
socgrp
|
-0.025
|
0.043
|
-0.59
|
0.558
|
-0.111
|
0.059
|
mpce_mrp
|
0.000
|
7.74e-07
|
24.07
|
0.000
|
0.000
|
0.000
|
Constant
|
-1.824
|
0.261
|
-7.00
|
0.000
|
-2.335
|
-1.313
|
group variable considered- socgrp
offset variable (considering fixed effect) – none, since the social group is group variable, the effect considered is random.
level- 1 (group variable- socgrp)
* specific dummy
*Monthly per capita expenditure (mpce), Type of district (district), Availability of internet (internet), Nature of liquidity (regsal), Fuel quality/calorific value of fuel for cooking for a household (cv)
Presence of social groups (stat reg), belonging to a particular community (rel) are the controls
Outcome Table 3
No. of Observation- 5000
No. of Groups-4
Integration Points- 7
Log likelihood - -2661.4289
Wald Chi2- 600.11
Prob> Chi2- 0.0000
Random-effects Parameters
|
Estimate
|
Standard Error
|
95% Confidence Interval
|
socgrp
|
0.253
|
0.102
|
0.114
|
0.560
|
LR test vs. logistic regression: chibar2(01) = 30.26 Prob>=chibar2 = 0.0000
Monthly per capita expenditure (mpce) , Monthly per capita expenditure for marginal rural population (mpce_mrp), Type of district (district), Availability of internet (internet)
Nature of liquidity (regsal), Household size (hh_size), Characteristics of households (hh_type) , Fuel quality/calorific value of fuel for cooking for a household (cv) , Presence of social groups (stat reg), Belonging to a particular community (rel) are all controls
Table 2: Impact of district profiles, presence of social groups and fuel quality on the choice of cooking fuels
|
Coefficient
|
Standard error
|
Z
|
P>|z|
|
cv
|
-5.09e-06
|
1.65e-06
|
-3.08
|
0.002
|
stat_reg
|
0.0064756
|
0.0013973
|
4.63
|
0.000
|
District
|
0.0207977
|
0.0084707
|
2.46
|
0.014
|
Constant
|
-7.572664
|
0.4672319
|
-16.21
|
0.000
|
** Monthly per capita expenditure (mpce), Monthly per capita expenditure for marginal rural population (mpce_mrp), Availability of internet (internet), Nature of liquidity (regsal)
Household size (hh_size), Characteristics of households (hh_type) , Belonging to a particular community (rel), Presence of social/cultural grouping or institutional network (socgrp) are all controls
Table 3: A Reliability Test of the Model Results
Total Population
|
63,061
|
Likelihood Ratio
|
13.96
|
Probability
|
0.0000009
|
Log value of the likelihood ratio
|
-37.053
|
Explainability Factor
|
0.16
|
**High likelihood ratio and low probability value of less than 5% and a positive explainability factor statistically indicates the significance and reliability of the model
Table 4: Impact of social groups, communities, household size on the choice of cooking
|
Coefficient
|
Standard Error
|
Z
|
P>|z|
|
cv
|
-0.0145812
|
0.0015198
|
-9.59
|
0.000
|
district
|
0.0152199
|
0.0073121
|
2.08
|
0.037
|
hh_size
|
0.3585109
|
0.0337609
|
10.62
|
0.000
|
rel*
|
0.2823244
|
0.1038493
|
2.72
|
0.007
|
socgrp**
|
0.1805408
|
0.0335194
|
5.39
|
0.000
|
Constant
|
-7.376072
|
0.3716588
|
-19.85
|
0.000
|
population size=63051
log likelihood Ratio Chi Square Statistics=16286.50
Probability of the likelihood=0.0000
Logarithmic value of likelihood of convergence = -35886
Ratio of Explainability=0.1850
*proxy of belonging to a particular community group/identity[1]
**proxy of belonging to local institutional frameworks / groups like self-help groups (SHGs)[2]
*** Monthly per capita expenditure (mpce) , monthly per capita expenditure for marginal rural population (mpce_mrp), availability of internet (internet), nature of liquidity (regsal), characteristics of households (hh_type) , presence of social groups (stat reg) are the controls
Table 5: Impact of select variables on choice of cooking fuel
|
Coefficient
|
Standard Error
|
Z
|
P>|z|
|
cv
|
-0.0079013
|
0.0005242
|
-15.07
|
0.000
|
mpce
|
2.81e-06
|
5.24e-07
|
-5.36
|
0.000
|
internet
|
0.0159915
|
0.4663565
|
0.03
|
0.973
|
regsal
|
0.449563
|
0.0950867
|
4.73
|
0.000
|
district
|
0.0692122
|
0.0016873
|
41.02
|
0.000
|
constant
|
-5.220253
|
0.9572833
|
-5.45
|
0.000
|
Multinomial logistic regression Number of observations=83362
LR chi2=2078.82
Prob > chi2=0.0000
Log likelihood = -5776.6252 Pseudo R2=0.1525
** Monthly per capita expenditure for marginal rural population (mpce_mrp)Type of district (district), Household size (hh_size), Characteristics of households (hh_type) , Presence of social groups (stat reg), Belonging to a particular community (rel), Presence of social/cultural grouping or institutional network (socgrp) are the controls
Table 6: State-level energy inequality index
|
State-level energy inequality index {A€ (E1, E2,. En)}
|
Bihar
|
Odisha
|
Primary fuel for cooking
|
0.38
|
0.33
|
LPG for cooking
|
0.28
|
0.38
|
Income
|
0.32
|
0.25
|