#########################################
#########################################
setwd("~/workspace/lits/attributions/")

#########################################
load("~/workspace/lits/attributions/data/lits.RData")
df.model <- df[df$group_general != "Western Europe",]

Description

This page includes everything related to modelling of the data. Click here for the final tables.

In this paper we have two depentend variables (DV) that are both binomial

support for social blame type of explanation
support for individual blame type of explanation

Unilevel regression analysis

Unilevel is used here for making distinction between multilevel analysis later. In the analysis of this chapter no contextual level variables are used.

I’m using the University of California’s resource R Data Analysis Examples: Logit Regression as a reference here.

Bivariate logistic regression

unbisoc.1 <- glm(pov.log.social ~ incsour3, data = df.model, family = "binomial")
unbisoc.2 <- glm(pov.log.social ~ edu2, data = df.model, family = "binomial")
unbisoc.3 <- glm(pov.log.social ~ income2, data = df.model, family = "binomial")
unbisoc.4 <- glm(pov.log.social ~ past.diff, data = df.model, family = "binomial")
unbisoc.5 <- glm(pov.log.social ~ future.diff, data = df.model, family = "binomial")
unbisoc.6 <- glm(pov.log.social ~ crise, data = df.model, family = "binomial")

library(texreg)

## Version:  1.30.2
## Date:     2013-12-10
## Author:   Philip Leifeld (University of Konstanz)


model.names = c("Transfer Dependency", "Education level",
                "Perceived income","Income compared to past",
                "Income compared to future","Effect of financial crises")

coef.names = c("(intercept)", 
               "Dependent vs. non-dependent",
               "Low education vs. non-low education",
               "Low income vs. high income",
               "Income has worsened vs. has not",
               "Income will worsen vs. will not",
               "Has affected great or fair amount vs. has not")

htmlreg(list(unbisoc.1,unbisoc.2,
             unbisoc.3,unbisoc.4,
             unbisoc.5,unbisoc.6),
        custom.model.names = model.names,
        custom.coef.names = coef.names,
        inline.css = FALSE,
        doctype = FALSE,
        html.tag = FALSE,
        head.tag = FALSE,
        body.tag = FALSE,
       caption="Simple logistic regression on social blame", caption.above=TRUE)

Simple logistic regression on social blame
	Transfer Dependency	Education level	Perceived income	Income compared to past	Income compared to future	Effect of financial crises
(intercept)	-0.20^***	-0.53^***	-0.12^***	-0.11^***	0.02	-0.07^***
	(0.06)	(0.12)	(0.02)	(0.02)	(0.03)	(0.02)
Dependent vs. non-dependent	-0.16^**
	(0.06)
Low education vs. non-low education		0.17
		(0.12)
Low income vs. high income			-0.41^***
			(0.03)
Income has worsened vs. has not				-0.39^***
				(0.03)
Income will worsen vs. will not					-0.49^***
					(0.04)
Has affected great or fair amount vs. has not						-0.46^***
						(0.03)
AIC	31792.27	31857.28	31633.41	30682.55	24863.69	29145.44
BIC	31808.40	31873.41	31649.54	30698.61	24879.34	29161.41
Log Likelihood	-15894.14	-15926.64	-15814.71	-15339.27	-12429.84	-14570.72
Deviance	31788.27	31853.28	31629.41	30678.55	24859.69	29141.44
Num. obs.	23461	23504	23513	22777	18526	21614
^*p < 0.001, ^p < 0.01, ^*p < 0.05

Interpretation

all the predictors are statistically significant which means that the null hypothesis of
- The probability of supporting the social blame type of explanation decreases by -0.16 if we compare individual who is dependent on tranfers with individual who is not dependent
- The probability of supporting the social blame type of explanation decreases by -0.08 if we compare individual with low education with individual with no-low education
- The probability of supporting the social blame type of explanation decreases by -0.59 if we compare individual with low perceived income with individual with high perceived income
- The probability of supporting the social blame type of explanation decreases by -0.39 if we compare individual whose income has worsened during the financial crisis with individual whose income has not worsened
- The probability of supporting the social blame type of explanation decreases by -0.49 if we compare individual who is expecting his income to worsen during the next four years with individual expects his income to remain at the same level or to increase
- The probability of supporting the social blame type of explanation decreases by -0.46 if we compare individual who has suffered Great or fair amount from the financial crisis with individual who has suffered Little or not at all
As you would expect the model fit is the best in the case of expected change in income

Explaining individual blame

unbiind.1 <- glm(pov.log.individual ~ incsour3, data = df.model, family = "binomial")
unbiind.2 <- glm(pov.log.individual ~ edu2, data = df.model, family = "binomial")
unbiind.3 <- glm(pov.log.individual ~ income2, data = df.model, family = "binomial")
unbiind.4 <- glm(pov.log.individual ~ past.diff, data = df.model, family = "binomial")
unbiind.5 <- glm(pov.log.individual ~ future.diff, data = df.model, family = "binomial")
unbiind.6 <- glm(pov.log.individual ~ crise, data = df.model, family = "binomial")

library(texreg)
htmlreg(list(unbiind.1,unbiind.2,
             unbiind.3,unbiind.4,
             unbiind.5,unbiind.6),
        custom.model.names = model.names,
        custom.coef.names = coef.names,
        inline.css = FALSE,
        doctype = FALSE,
        html.tag = FALSE,
        head.tag = FALSE,
        body.tag = FALSE,
       caption="Simple logistic regression on individual blame", caption.above=TRUE)

Simple logistic regression on individual blame
	Transfer Dependency	Education level	Perceived income	Income compared to past	Income compared to future	Effect of financial crises
(intercept)	-1.67^***	-1.20^***	-1.41^***	-1.41^***	-1.48^***	-1.52^***
	(0.08)	(0.14)	(0.03)	(0.03)	(0.04)	(0.03)
Dependent vs. non-dependent	0.49^***
	(0.08)
Low education vs. non-low education		0.00
		(0.14)
Low income vs. high income			0.35^***
			(0.03)
Income has worsened vs. has not				0.32^***
				(0.03)
Income will worsen vs. will not					0.40^***
					(0.05)
Has affected great or fair amount vs. has not						0.53^***
						(0.03)
AIC	25314.01	25393.10	25281.83	24576.83	20388.41	23070.37
BIC	25330.13	25409.23	25297.96	24592.90	20404.06	23086.33
Log Likelihood	-12655.00	-12694.55	-12638.92	-12286.42	-10192.20	-11533.19
Deviance	25310.01	25389.10	25277.83	24572.83	20384.41	23066.37
Num. obs.	23461	23504	23513	22777	18526	21614
^*p < 0.001, ^p < 0.01, ^*p < 0.05

Interpretation

all the predictors, but education level are statistically significant
- The probability of supporting the individual blame type of explanation increases by 0.49 if we compare individual who is dependent on tranfers to individual who is not dependent
- The probability of supporting the individual blame type of explanation increases by 0.03 if we compare individual with low education with individual with no-low education. This is not statistically significant
- The probability of supporting the individual blame type of explanation increases by 0.38 if we compare individual with low perceived income with individual with high perceived income
- The probability of supporting the individual blame type of explanation increases by 0.32 if we compare individual whose income has worsened during the financial crisis with individual whose income has not worsened
- The probability of supporting the individual blame type of explanation increases by 0.40 if we compare individual who is expecting his income to worsen during the next four years with individual expects his income to remain at the same level or to increase
- The probability of supporting the individual blame type of explanation increases by -0.53 if we compare individual who has suffered Great or fair amount from the financial crisis with individual who has suffered Little or not at all
As you would expect the model fit is the best in the case of expected change in income

Multivariate logistic regression

mulbisoc.1 <- glm(pov.log.social ~  incsour3, data = df.model, family = "binomial")
mulbisoc.2 <- glm(pov.log.social ~  incsour3 +
                                    edu2, data = df.model, family = "binomial")
mulbisoc.3 <- glm(pov.log.social ~  incsour3 +
                                    edu2 +
                                    income2, data = df.model, family = "binomial")
mulbisoc.4 <- glm(pov.log.social ~  incsour3 +
                                    edu2 +
                                    income2 +
                                    past.diff, data = df.model, family = "binomial")
mulbisoc.4 <- glm(pov.log.social ~  incsour3 +
                                    edu2 +
                                    income2 +
                                    past.diff, data = df.model, family = "binomial")
mulbisoc.5 <- glm(pov.log.social ~  incsour3 +
                                    edu2 +
                                    income2 +
                                    past.diff + 
                                    future.diff, data = df.model, family = "binomial")
mulbisoc.6 <- glm(pov.log.social ~  incsour3 +
                                    edu2 +
                                    income2 +
                                    past.diff + 
                                    future.diff +
                                    crise, data = df.model, family = "binomial")

library(texreg)
htmlreg(list(mulbisoc.1,mulbisoc.2,
             mulbisoc.3,mulbisoc.4,
             mulbisoc.5,mulbisoc.6),
        custom.coef.names = coef.names,
        inline.css = FALSE,
        doctype = FALSE,
        html.tag = FALSE,
        head.tag = FALSE,
        body.tag = FALSE,
       caption="Multiple logistic regression on social blame", caption.above=TRUE)

Multiple logistic regression on social blame
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
(intercept)	-0.20^***	-0.38^**	-0.28^*	-0.04	0.42^**	0.50^**
	(0.06)	(0.13)	(0.13)	(0.14)	(0.16)	(0.17)
Dependent vs. non-dependent	-0.16^**	-0.16^**	-0.13^*	-0.11	-0.17^*	-0.14
	(0.06)	(0.06)	(0.06)	(0.06)	(0.07)	(0.07)
Low education vs. non-low education		0.18	0.29^*	0.23	0.12	0.12
		(0.12)	(0.12)	(0.12)	(0.14)	(0.15)
Low income vs. high income			-0.42^***	-0.39^***	-0.40^***	-0.37^***
			(0.03)	(0.03)	(0.03)	(0.03)
Income has worsened vs. has not				-0.32^***	-0.25^***	-0.17^***
				(0.03)	(0.03)	(0.03)
Income will worsen vs. will not					-0.44^***	-0.40^***
					(0.04)	(0.04)
Has affected great or fair amount vs. has not						-0.36^***
						(0.03)
AIC	31792.27	31779.53	31542.26	30413.96	24393.11	22663.10
BIC	31808.40	31803.72	31574.51	30454.12	24440.02	22717.32
Log Likelihood	-15894.14	-15886.76	-15767.13	-15201.98	-12190.55	-11324.55
Deviance	31788.27	31773.53	31534.26	30403.96	24381.11	22649.10
Num. obs.	23461	23452	23452	22721	18374	17089
^*p < 0.001, ^p < 0.01, ^*p < 0.05

Interpretation

all the predictors, but education level are statistically significant
- The probability of supporting the individual blame type of explanation increases by 0.49 if we compare individual who is dependent on tranfers to individual who is not dependent
- The probability of supporting the individual blame type of explanation increases by 0.03 if we compare individual with low education with individual with no-low education. This is not statistically significant
- The probability of supporting the individual blame type of explanation increases by XXX` if we compare individual with low perceived income with individual with high perceived income
- The probability of supporting the individual blame type of explanation increases by 0.32 if we compare individual whose income has worsened during the financial crisis with individual whose income has not worsened
- The probability of supporting the individual blame type of explanation increases by 0.40 if we compare individual who is expecting his income to worsen during the next four years with individual expects his income to remain at the same level or to increase
- The probability of supporting the individual blame type of explanation increases by -XXX if we compare individual who has suffered Great or fair amount from the financial crisis with individual who has suffered Little or not at all
As you would expect the model fit is the best in the case of expected change in income

Individual blame

mulbiind.1 <- glm(pov.log.individual ~  incsour3, data = df.model, family = "binomial")
mulbiind.2 <- glm(pov.log.individual ~  incsour3 +
                                    edu2, data = df.model, family = "binomial")
mulbiind.3 <- glm(pov.log.individual ~  incsour3 +
                                    edu2 +
                                    income2, data = df.model, family = "binomial")
mulbiind.4 <- glm(pov.log.individual ~  incsour3 +
                                    edu2 +
                                    income2 +
                                    past.diff, data = df.model, family = "binomial")
mulbiind.4 <- glm(pov.log.individual ~  incsour3 +
                                    edu2 +
                                    income2 +
                                    past.diff, data = df.model, family = "binomial")
mulbiind.5 <- glm(pov.log.individual ~  incsour3 +
                                    edu2 +
                                    income2 +
                                    past.diff + 
                                    future.diff, data = df.model, family = "binomial")
mulbiind.6 <- glm(pov.log.individual ~  incsour3 +
                                    edu2 +
                                    income2 +
                                    past.diff + 
                                    future.diff +
                                    crise, data = df.model, family = "binomial")

library(texreg)
htmlreg(list(mulbiind.1,mulbiind.2,
             mulbiind.3,mulbiind.4,
             mulbiind.5,mulbiind.6),
        custom.coef.names = coef.names,
        inline.css = FALSE,
        doctype = FALSE,
        html.tag = FALSE,
        head.tag = FALSE,
        body.tag = FALSE,
       caption="Multiple logistic regression on individual blame", caption.above=TRUE)

Multiple logistic regression on individual blame
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
(intercept)	-1.67^***	-1.66^***	-1.76^***	-1.93^***	-2.06^***	-2.21^***
	(0.08)	(0.16)	(0.16)	(0.16)	(0.19)	(0.20)
Dependent vs. non-dependent	0.49^***	0.49^***	0.46^***	0.45^***	0.39^***	0.33^***
	(0.08)	(0.08)	(0.08)	(0.08)	(0.09)	(0.09)
Low education vs. non-low education		0.00	-0.09	-0.06	-0.08	-0.05
		(0.14)	(0.14)	(0.14)	(0.16)	(0.18)
Low income vs. high income			0.34^***	0.30^***	0.29^***	0.26^***
			(0.03)	(0.03)	(0.04)	(0.04)
Income has worsened vs. has not				0.27^***	0.21^***	0.14^***
				(0.03)	(0.04)	(0.04)
Income will worsen vs. will not					0.36^***	0.33^***
					(0.05)	(0.05)
Has affected great or fair amount vs. has not						0.44^***
						(0.04)
AIC	25314.01	25308.83	25198.07	24417.39	20104.00	18521.25
BIC	25330.13	25333.01	25230.32	24457.54	20150.91	18575.47
Log Likelihood	-12655.00	-12651.41	-12595.03	-12203.69	-10046.00	-9253.62
Deviance	25310.01	25302.83	25190.07	24407.39	20092.00	18507.25
Num. obs.	23461	23452	23452	22721	18374	17089
^*p < 0.001, ^p < 0.01, ^*p < 0.05

Interpretation

Multinomial logistic regression

In multinomial setting we use variable povertyas DV. Dont know and not stated values are exluded from the analysis and we are left with following four values:

Social Blame
Individual Blame
Individual Fate
Social Fate

Of the DV values we use Social Blame as a reference category as it is distinctive category and allows me to make clear interpretations of the significance of the predictors. (source)

Simple multinomial logistic regression

using mlogit-package

This follows the example from University of Toronto and Discrete-Choice Logit Models with R by Philip A. Viton (October 26, 2013)

library(mlogit)

## Loading required package: Formula
## Loading required package: maxLik
## Loading required package: miscTools
## 
## Please cite the 'maxLik' package as:
## Henningsen, Arne and Toomet, Ott (2011). maxLik: A package for maximum likelihood estimation in R. Computational Statistics 26(3), 443-458. DOI 10.1007/s00180-010-0217-1.
## 
## If you have questions, suggestions, or comments regarding the 'maxLik' package, please use a forum or 'tracker' at maxLik's R-Forge site:
## https://r-forge.r-project.org/projects/maxlik/

dat.mlogit <- df.model[,c("poverty","edu2","incsour3","income2",
                          "past.diff","future.diff","crise")]
 dat.mlogit <- mlogit.data(dat.mlogit, shape="wide", choice="poverty")
# ------- #
model.mlogit5 <- mlogit(poverty ~ 0 | incsour3 + edu2 + 
                                      past.diff + future.diff +
                                      crise, 
                               data = dat.mlogit, 
                               reflevel ="Social Blame")

summary(model.mlogit5)

## 
## Call:
## mlogit(formula = poverty ~ 0 | incsour3 + edu2 + past.diff + 
##     future.diff + crise, data = dat.mlogit, reflevel = "Social Blame", 
##     method = "nr", print.level = 0)
## 
## Frequencies of alternatives:
##     Social Blame Individual Blame  Individual Fate      Social Fate 
##            0.450            0.262            0.110            0.178 
## 
## nr method
## 5 iterations, 0h:0m:2s 
## g'(-H)^-1g = 3.84E-05 
## successive function values within tolerance limits 
## 
## Coefficients :
##                                             Estimate Std. Error t-value
## Individual Blame:(intercept)               -1.777836   0.220002   -8.08
## Individual Fate:(intercept)                -1.946130   0.265878   -7.32
## Social Fate:(intercept)                    -1.764688   0.269663   -6.54
## Individual Blame:incsour3notDependent       0.372283   0.099007    3.76
## Individual Fate:incsour3notDependent        0.011231   0.120737    0.09
## Social Fate:incsour3notDependent            0.027272   0.099785    0.27
## Individual Blame:edu2higher                 0.000904   0.193010    0.00
## Individual Fate:edu2higher                 -0.349059   0.230661   -1.51
## Social Fate:edu2higher                      0.365868   0.248106    1.47
## Individual Blame:past.diffsame or better    0.270862   0.043845    6.18
## Individual Fate:past.diffsame or better     0.228006   0.059970    3.80
## Social Fate:past.diffsame or better         0.159721   0.048618    3.29
## Individual Blame:future.diffsame or better  0.450845   0.053514    8.42
## Individual Fate:future.diffsame or better   0.524579   0.076319    6.87
## Social Fate:future.diffsame or better       0.269378   0.057731    4.67
## Individual Blame:criseLittle or not at all  0.587613   0.041704   14.09
## Individual Fate:criseLittle or not at all   0.518815   0.057026    9.10
## Social Fate:criseLittle or not at all       0.258517   0.046222    5.59
##                                            Pr(>|t|)    
## Individual Blame:(intercept)                6.7e-16 ***
## Individual Fate:(intercept)                 2.5e-13 ***
## Social Fate:(intercept)                     6.0e-11 ***
## Individual Blame:incsour3notDependent       0.00017 ***
## Individual Fate:incsour3notDependent        0.92588    
## Social Fate:incsour3notDependent            0.78462    
## Individual Blame:edu2higher                 0.99626    
## Individual Fate:edu2higher                  0.13020    
## Social Fate:edu2higher                      0.14031    
## Individual Blame:past.diffsame or better    6.5e-10 ***
## Individual Fate:past.diffsame or better     0.00014 ***
## Social Fate:past.diffsame or better         0.00102 ** 
## Individual Blame:future.diffsame or better  < 2e-16 ***
## Individual Fate:future.diffsame or better   6.3e-12 ***
## Social Fate:future.diffsame or better       3.1e-06 ***
## Individual Blame:criseLittle or not at all  < 2e-16 ***
## Individual Fate:criseLittle or not at all   < 2e-16 ***
## Social Fate:criseLittle or not at all       2.2e-08 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Log-Likelihood: -19500
## McFadden R^2:  0.0137 
## Likelihood ratio test : chisq = 543 (p.value = <2e-16)

Interpretation

Multilevel multinomial logistic regression

Can be modelled in R in Bayesian manner using either ChoiceModelR and Zelig.

Marco Steenbergen discusses the stat theory and imlementation in Stata in this video

Multilevel regression

Bivariate analysis

library(lme4)

## Loading required package: lattice
## Loading required package: Matrix

df.model$group_balinCis[df.model$cntry %in% c('Czech Republic',
                                        'Hungary','Bulgaria',
                                        'Poland','Slovakia',
                                        'Slovenia','Romania')] <- 'CEE'

df.model$group_balinCis[df.model$cntry %in% c('Armenia','Azerbaijan',
                                        'Belarus','Georgia',
                                        'Kazakhstan','Kyrgyzstan',
                                        'Moldova','Tajikistan',
                                        'Ukraine','Uzbekistan','Russia',
                                        'Estonia','Latvia','Lithuania')] <- 'CIS'
df.model$group_balinCis <- factor(df.model$group_balinCis, levels=c("CEE","CIS"))


bisoc.1 <- glmer(pov.log.social ~ incsour3+(1|cntry),
                        family = binomial(logit), data = df.model)
bisoc.2 <- glmer(pov.log.social ~ edu2+(1|cntry),
                        family = binomial(logit), data = df.model)
bisoc.3 <- glmer(pov.log.social ~ income2+(1|cntry),
                        family = binomial(logit), data = df.model)
bisoc.4 <- glmer(pov.log.social ~ past.diff+(1|cntry),
                        family = binomial(logit), data = df.model)
bisoc.5 <- glmer(pov.log.social ~ future.diff+(1|cntry),
                        family = binomial(logit), data = df.model)
bisoc.6 <- glmer(pov.log.social ~ crise+(1|cntry),
                         family = binomial(logit), data = df.model)
bisoc.8 <- glmer(pov.log.social ~ undp_hdi+(1|cntry),
                         family = binomial(logit), data = df.model)
bisoc.9 <- glmer(pov.log.social ~ gdpChange+(1|cntry),
                         family = binomial(logit), data = df.model)
bisoc.10 <- glmer(pov.log.social ~ wbgi_vae+(1|cntry),
                          family = binomial(logit), data = df.model)
bisoc.11 <- glmer(pov.log.social ~ group_analysis+(1|cntry),
                          family = binomial(logit), data = df.model)
bisoc.12 <- glmer(pov.log.social ~ group_balinCis+(1|cntry),
                          family = binomial(logit), data = df.model)
bisoc.13 <- glmer(pov.log.social ~ perGini+(1|cntry),
                          family = binomial(logit), data = df.model)


coef.names.multi = c("(intercept)", 
                      "Dependent","Low education",
                      "Low income","Income has worsened",
                      "Income will worsen","Has affected great or fair amount",
                      "Human Development Index",
                      "Change in GDP between 2007 to 2010",
                      "Good governance","Cgroup","CgroupMod",
                     "Perceived Gini")

library(texreg)
htmlreg(list(bisoc.1,bisoc.2,
          bisoc.3,bisoc.4,
          bisoc.5,bisoc.6,
          bisoc.8,
          bisoc.9,bisoc.10,
          bisoc.11,bisoc.12,
          bisoc.13),
        #custom.coef.names = coef.names.multi,
        naive = TRUE,
        inline.css = FALSE,
        doctype = FALSE,
        html.tag = FALSE,
        head.tag = FALSE,
        body.tag = FALSE,
        include.pvalues=TRUE,
        caption="Bivariate social blame", caption.above=TRUE)

## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.

Bivariate social blame
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8	Model 9	Model 10	Model 11	Model 12
(Intercept)	-0.19	-0.55^**	-0.16	-0.23	-0.12	-0.14	-2.34^*	-0.17	-0.36^***	-0.26	-0.34	-2.01^**
	(0.13)	(0.17)	(0.12)	(0.12)	(0.12)	(0.11)	(1.06)	(0.11)	(0.11)	(0.17)	(0.20)	(0.63)
incsour3notDependent	-0.21^***
	(0.06)
edu2higher		0.16
		(0.12)
income2High			-0.39^***
			(0.03)
past.diffsame or better				-0.25^***
				(0.03)
future.diffsame or better					-0.37^***
					(0.04)
criseLittle or not at all						-0.42^***
						(0.03)
undp_hdi							2.59
							(1.41)
gdpChange								-0.01^***
								(0.00)
wbgi_vae									0.23^*
									(0.11)
group_analysisCIS										-0.24
										(0.23)
group_balinCisCIS											-0.07
											(0.25)
perGini												0.08^**
												(0.03)
AIC	30262.21	30331.32	30147.92	29305.05	23667.34	27981.03	30342.24	30334.30	30340.93	30344.31	30345.33	30339.48
BIC	30286.40	30355.52	30172.11	29329.15	23690.82	28004.97	30366.44	30358.50	30365.13	30368.50	30369.52	30363.68
Log Likelihood	-15128.11	-15162.66	-15070.96	-14649.53	-11830.67	-13987.51	-15168.12	-15164.15	-15167.47	-15169.15	-15169.66	-15166.74
Deviance	30256.21	30325.32	30141.92	29299.05	23661.34	27975.03	30336.24	30328.30	30334.93	30338.31	30339.33	30333.48
Num. obs.	23461	23504	23513	22777	18526	21614	23513	23513	23513	23513	23513	23513
Num. groups: cntry	21	21	21	21	21	21	21	21	21	21	21	21
Variance: cntry.(Intercept)	0.29	0.29	0.29	0.28	0.29	0.26	0.25	0.17	0.23	0.27	0.29	0.22
Variance: Residual	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00
^*p < 0.001, ^p < 0.01, ^*p < 0.05

Interpretation

Individual blame explained

biind.1 <- glmer(pov.log.individual ~ incsour3+(1|cntry),
                        family = binomial(logit), data = df.model)
biind.2 <- glmer(pov.log.individual ~ edu2+(1|cntry),
                        family = binomial(logit), data = df.model)
biind.3 <- glmer(pov.log.individual ~ income2+(1|cntry),
                        family = binomial(logit), data = df.model)
biind.4 <- glmer(pov.log.individual ~ past.diff+(1|cntry),
                        family = binomial(logit), data = df.model)
biind.5 <- glmer(pov.log.individual ~ future.diff+(1|cntry),
                        family = binomial(logit), data = df.model)
biind.6 <- glmer(pov.log.individual ~ crise+(1|cntry),
                        family = binomial(logit), data = df.model)
biind.8 <- glmer(pov.log.individual ~ undp_hdi+(1|cntry),
                        family = binomial(logit), data = df.model)
biind.9 <- glmer(pov.log.individual ~ gdpChange+(1|cntry),
                        family = binomial(logit), data = df.model)
biind.10 <- glmer(pov.log.individual ~ wbgi_vae+(1|cntry),
                         family = binomial(logit), data = df.model)
biind.11 <- glmer(pov.log.individual ~ group_analysis+(1|cntry),
                          family = binomial(logit), data = df.model)
biind.12 <- glmer(pov.log.individual ~ group_balinCis+(1|cntry),
                          family = binomial(logit), data = df.model)
biind.13 <- glmer(pov.log.individual ~ perGini+(1|cntry),
                          family = binomial(logit), data = df.model)

library(texreg)
htmlreg(list(biind.1,biind.2,
          biind.3,biind.4,
          biind.5,biind.6,
          biind.8,
          biind.9,biind.10,
          biind.11,biind.12,
             biind.13),
        #custom.coef.names = coef.names.multi,
        naive = TRUE,
        inline.css = FALSE,
        doctype = FALSE,
        html.tag = FALSE,
        head.tag = FALSE,
        body.tag = FALSE,
        use.packages=FALSE,
        include.pvalues=TRUE,
        caption="Bivariate individual blame", 
        caption.above=TRUE)

## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.

Bivariate individual blame
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8	Model 9	Model 10	Model 11	Model 12
(Intercept)	-1.72^***	-1.23^***	-1.47^***	-1.40^***	-1.44^***	-1.52^***	0.39	-1.40^***	-1.26^***	-1.33^***	-1.28^***	-0.42
	(0.11)	(0.16)	(0.08)	(0.08)	(0.08)	(0.08)	(0.65)	(0.07)	(0.07)	(0.11)	(0.13)	(0.43)
incsour3notDependent	0.49^***
	(0.08)
edu2higher		-0.02
		(0.14)
income2High			0.36^***
			(0.03)
past.diffsame or better				0.24^***
				(0.04)
future.diffsame or better					0.31^***
					(0.05)
criseLittle or not at all						0.50^***
						(0.04)
undp_hdi							-2.18^*
							(0.86)
gdpChange								0.01^***
								(0.00)
wbgi_vae									-0.16^*
									(0.07)
group_analysisCIS										0.16
										(0.15)
group_balinCisCIS											0.04
											(0.16)
perGini												-0.04^*
												(0.02)
AIC	24816.34	24895.46	24777.07	24131.55	20052.68	22693.65	24896.35	24888.58	24897.13	24900.86	24901.94	24898.49
BIC	24840.53	24919.65	24801.27	24155.66	20076.16	22717.59	24920.55	24912.78	24921.32	24925.06	24926.14	24922.68
Log Likelihood	-12405.17	-12444.73	-12385.54	-12062.78	-10023.34	-11343.82	-12445.18	-12441.29	-12445.56	-12447.43	-12447.97	-12446.24
Deviance	24810.34	24889.46	24771.07	24125.55	20046.68	22687.65	24890.35	24882.58	24891.13	24894.86	24895.94	24892.49
Num. obs.	23461	23504	23513	22777	18526	21614	23513	23513	23513	23513	23513	23513
Num. groups: cntry	21	21	21	21	21	21	21	21	21	21	21	21
Variance: cntry.(Intercept)	0.12	0.12	0.12	0.11	0.10	0.11	0.09	0.06	0.09	0.11	0.12	0.10
Variance: Residual	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00
^*p < 0.001, ^p < 0.01, ^*p < 0.05

Interpretation

Multivariate multilevel regressions

library(lme4)
soc.0 <- glmer(pov.log.social ~ (1|cntry),
                      family = binomial(logit),
                       data = df.model)
soc.1 <- glmer(pov.log.social ~ incsour3+edu2+income2+
                        past.diff+future.diff+crise+(1|cntry),
                      family = binomial(logit),
                      data = df.model)
soc.2 <- glmer(pov.log.social ~ incsour3+edu2+income2+
                        past.diff+future.diff+crise+
                        undp_hdi+(1|cntry),
                       family = binomial(logit),
                       data = df.model)
soc.3 <- glmer(pov.log.social ~ incsour3+edu2+income2+
                        past.diff+future.diff+crise+
                        undp_hdi+gdpChange+(1|cntry),
                       family = binomial(logit),
                       data = df.model)
soc.4 <- glmer(pov.log.social ~ incsour3+edu2+income2+
                        past.diff+future.diff+crise+
                        undp_hdi+gdpChange+wbgi_vae+(1|cntry),
                       family = binomial(logit),
                       data = df.model)
soc.Cont <- glmer(pov.log.social ~ undp_hdi+gdpChange+
                      wbgi_vae+(1|cntry),
                       family = binomial(logit),
                       data = df.model)

coef.names.multi = c("(intercept)", 
                      "Dependent","Low education",
                      "Low income","Income has worsened",
                      "Income will worsen","Has affected great or fair amount",
                      "Human Deveplopment Index",
                      "Change in GDP between 2007 to 2010",
                      "Good governance")


model.names.multi = c("Empty",
                      "Only individual",
                      "HDI added",
                      "GDP change added",
                      "Good governance added")

htmlreg(list(soc.0,
        soc.1,
        soc.2,
        soc.3,
        soc.4),
        custom.model.names = model.names.multi,
        custom.coef.names = coef.names.multi,
        naive = TRUE,
        inline.css = FALSE,
        doctype = FALSE,
        html.tag = FALSE,
        head.tag = FALSE,
        body.tag = FALSE,
        include.pvalues=TRUE,
        caption="Logistic multilevel random intercept model for social blame type of explanation", 
        caption.above=TRUE)

Interpretation

Individual blame type of explanation

ind.0 <- glmer(pov.log.individual ~ (1|cntry),
                      family = binomial(logit),
                       data = df.model)
ind.1 <- glmer(pov.log.individual ~ incsour3+edu2+income2+
                        past.diff+future.diff+crise+(1|cntry),
                      family = binomial(logit),
                       data = df.model)
ind.2 <- glmer(pov.log.individual ~ incsour3+edu2+income2+
                        past.diff+future.diff+crise+
                        undp_hdi+(1|cntry),
                       family = binomial(logit),
                       data = df.model)
ind.3 <- glmer(pov.log.individual ~ incsour3+edu2+income2+
                        past.diff+future.diff+crise+
                        undp_hdi+gdpChange+(1|cntry),
                       family = binomial(logit),
                       data = df.model)
ind.4 <- glmer(pov.log.individual ~ incsour3+edu2+income2+
                        past.diff+future.diff+crise+
                        undp_hdi+gdpChange+wbgi_vae+(1|cntry),
                       family = binomial(logit),
                       data = df.model)
ind.Cont <- glmer(pov.log.individual ~ undp_hdi+gdpChange+
                      wbgi_vae+(1|cntry),
                       family = binomial(logit),
                       data = df.model)

htmlreg(list(ind.0,
       ind.1,
       ind.2,
       ind.3,
       ind.4),
        custom.model.names = model.names.multi,
        custom.coef.names = coef.names.multi,
        naive = TRUE,
        ci.test = NULL,
        inline.css = FALSE,
        doctype = FALSE,
        html.tag = FALSE,
        head.tag = FALSE,
        body.tag = FALSE,
        include.pvalues=TRUE,
        caption="Logistic multilevel random intercept model for individual blame type of explanation", 
        caption.above=TRUE)

Comparison of full models with our last model



df.model$pov.log.socVsInc[df.model$poverty == "Social Blame"] <- 1
df.model$pov.log.socVsInc[df.model$poverty == "Individual Blame"] <- 0

socVsInc <- glmer(pov.log.socVsInc ~ incsour3+edu2+income2+
                        past.diff+future.diff+crise+
                        undp_hdi+gdpChange+wbgi_vae+(1|cntry),
                       family = binomial(logit),
                       data = df.model)

save(ind.0,ind.1,ind.2,ind.3,ind.4,ind.Cont,
     soc.0,soc.1,soc.2,soc.3,soc.4,soc.Cont,
     socVsInc, file="data/finalModels.RData")

Tables for presentation

Partial

FINAL TABLES SAME AS IN PAPER

library(texreg)

#load("data/finalModels.RData")

coef.names.multi = c("(intercept)", 
                      "Dependent","Low education",
                      "Low income","Income has worsened",
                      "Income will worsen","Has affected great or fair amount",
                      "Human Deveplopment Index",
                      "Change in GDP between 2007 to 2010",
                      "Voice and Accountability")

model.names.multi = c("Empty",
                      "Only individual",
                      "HDI added",
                      "GDP change added",
                      "Voice and Accountability added")

model.names.multi.A = c("SB empty",
                      "SB individual",
                      "SB contextual",
                      "SB all",
                      "IB empty",
                      "IB individual",
                      "IB contextual",
                      "IB all",
                      "SB vs. IB")

htmlreg(list(soc.0,soc.1,soc.Cont,soc.4,
            ind.0,ind.1,ind.Cont,ind.4,socVsInc),
        custom.model.names = model.names.multi.A,
        custom.coef.names = coef.names.multi,
        naive=TRUE,
        inline.css = TRUE,
        doctype = FALSE,
        html.tag = FALSE,
        head.tag = FALSE,
        body.tag = FALSE,
        include.pvalues=TRUE,
        caption="Logistic multilevel random intercept model for social and individual blame type of explanation", 
        caption.above=TRUE)

## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.
## confint.merMod method not found. Using naive p values instead.

Logistic multilevel random intercept model for social and individual blame type of explanation
	SB empty	SB individual	SB contextual	SB all	IB empty	IB individual	IB contextual	IB all	SB vs. IB
(intercept)	-0.39^**	0.36	0.38	0.88	-1.25^***	-2.19^***	-0.85	-1.24	1.61
	(0.12)	(0.21)	(1.51)	(1.48)	(0.08)	(0.22)	(0.90)	(0.86)	(1.34)
Dependent		-0.20^**		-0.20^**		0.37^***		0.37^***	-0.43^***
		(0.07)		(0.07)		(0.09)		(0.09)	(0.10)
Low education		0.11		0.11		-0.04		-0.04	-0.01
		(0.16)		(0.16)		(0.18)		(0.18)	(0.20)
Low income		-0.36^***		-0.37^***		0.28^***		0.28^***	-0.43^***
		(0.03)		(0.03)		(0.04)		(0.04)	(0.04)
Income has worsened		-0.06		-0.06		0.08		0.07	-0.10^*
		(0.04)		(0.04)		(0.04)		(0.04)	(0.05)
Income will worsen		-0.32^***		-0.32^***		0.27^***		0.26^***	-0.41^***
		(0.04)		(0.04)		(0.05)		(0.05)	(0.06)
Has affected great or fair amount		-0.32^***		-0.32^***		0.44^***		0.44^***	-0.53^***
		(0.04)		(0.04)		(0.04)		(0.04)	(0.05)
Human Deveplopment Index			-0.65	-0.36			-0.77	-1.45	0.73
			(1.96)	(1.91)			(1.17)	(1.09)	(1.72)
Change in GDP between 2007 to 2010			-0.02^**	-0.02^**			0.01^***	0.01^**	-0.02^***
			(0.01)	(0.01)			(0.00)	(0.00)	(0.01)
Voice and Accountability			-0.08	-0.08			0.13	0.13	-0.14
			(0.18)	(0.17)			(0.11)	(0.10)	(0.16)
AIC	30343.40	21780.91	30337.63	21776.53	24900.01	18249.54	24891.15	18239.74	13565.27
BIC	30359.53	21842.88	30377.96	21861.73	24916.15	18311.51	24931.47	18324.95	13645.80
Log Likelihood	-15169.70	-10882.45	-15163.82	-10877.26	-12448.01	-9116.77	-12440.57	-9108.87	-6771.63
Deviance	30339.40	21764.91	30327.63	21754.53	24896.01	18233.54	24881.15	18217.74	13543.27
Num. obs.	23513	17089	23513	17089	23513	17089	23513	17089	11169
Num. groups: cntry	21	21	21	21	21	21	21	21	21
Variance: cntry.(Intercept)	0.29	0.26	0.16	0.15	0.12	0.10	0.05	0.04	0.12
Variance: Residual	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00
^*p < 0.001, ^p < 0.01, ^*p < 0.05

Comparison of full models

coef.names.multi = c("(intercept)", 
                      "Dependent","Low education",
                      "Low income","Income has worsened",
                      "Income will worsen","Has affected great or fair amount",
                      "Human Development Index",
                      "Change in GDP between 2007 to 2010",
                      "Good governance")

model.names.multi.2 = c("Social blame vs. not",
                      "Individual blame vs. not",
                        "Social blame vs. Individual blame")

library(texreg)
htmlreg(list(soc.4,ind.4,socVsInc),
        naive = TRUE,
        custom.model.names = model.names.multi.2,
        custom.coef.names = coef.names.multi,
        inline.css = FALSE,
        doctype = FALSE,
        html.tag = FALSE,
        head.tag = FALSE,
        body.tag = FALSE,
        include.pvalues=TRUE,
        caption="Logistic multilevel random intercept model for social and individual blame type of explanation", 
        caption.above=TRUE)

Regression modellining

Description

Unilevel regression analysis

Bivariate logistic regression

Explaining social blame

Interpretation

Explaining individual blame

Interpretation

Multivariate logistic regression

Social blame

Interpretation

Individual blame

Interpretation

Multinomial logistic regression

Simple multinomial logistic regression

using mlogit-package

Interpretation

Multilevel multinomial logistic regression

Multilevel regression

Bivariate analysis

Social blame explained

Interpretation

Individual blame explained

Interpretation

Multivariate multilevel regressions

Social blame type of explanation

Interpretation

Individual blame type of explanation

Comparison of full models with our last model

Tables for presentation

Partial

FINAL TABLES SAME AS IN PAPER

Comparison of full models