How are Structural Inequalities Linked to the Wellbeing of Disabled Women in Costa Rica?
Erika Sanborne, University of Minnesota
I shared a preliminary iteration of this study at: Population Association of America (PAA 2023) Annual Meeting, New Orleans, LA US
Session: Health, Health Behaviors, and Healthcare
If anything here is not accessible to you, please let me know how I can help you.
Abstract
This study explores the relationships between structural inequalities and wellbeing for disabled women in Costa Rica. Analyzing a nationally representative sample of women from IPUMS MICS Round 6 (n=7,502), it examines life evaluation, measured by the Cantril ladder, as a proxy for subjective wellbeing.
Employing an integrative theoretical framework, this study combines the social stress process model, which explains how an array of social stressors and mediating social and personal resources affect wellbeing, with the capability approach, which considers how these stressors may collectively limit disabled women’s freedom to achieve the quality of life they desire.
The first hypothesis posits that disabled women of color in more disadvantaged regions will have lower subjective wellbeing compared to their non-disabled neighbors, indicating a main effect for disability. The second hypothesis suggests a main effect for structural inequality, predicting that disabled women’s subjective wellbeing will vary by ethnicity and subnational region.
Findings highlight the intersectionality of ethnicity, region, and disability, revealing important structural factors associated with women’s wellbeing in Costa Rica. Significant wellbeing gaps suggest that Black/Afro-Costa Rican and Indigenous disabled women are disproportionately left behind in the country’s sustainable development agenda.
Research Question
How are structural inequalities, such as ethnicity and region, linked to the wellbeing of disabled women in Costa Rica?
Background
Case Study: Costa Rica
When subjective wellbeing data are aggregated at the country level and appropriately weighted, Costa Rica emerges as a standout. From my preliminary, exploratory analysis of life evaluation using IPUMS MICS-6 data, Costa Rica emerged as a necessary case study.
Numerous other sources, such as the Happy Planet Index (Abdallah, Hoffman and Akenji 2024), the Sustainable Development Index (Hickel 2020), and the World Happiness Report (Helliwell et al. 2024), affirm Costa Rica’s comparatively highly ranked subjective wellbeing among countries, based on its high aggregate evaluative wellbeing scores in their measures as well.
These observations align with other assessments. The World Bank (2024) recognizes Costa Rica as “a success story in terms of development” due to its steady economic growth, stable democracy, and commitment to social progress. The Costa Rican economy is one of the strongest, and its poverty rates are among the lowest in the Latin America and Caribbean (LAC) region (World Bank 2024). By 2023, poverty levels had returned to pre-pandemic benchmarks.
The previous graphic gives a snapshot overview of where Costa Rica was with respect to their sustainable development goals as of 2023. While this summary reflects moderately improving amidst significant challenges for SDG #3, the detailed report reveals that their target for subjective wellbeing is “in the green” at 7.1. This suggests Costa Rica is on track towards the 2030 goal of 7.6. Costa Rica has also shown a commitment to expanding capabilities through investments in universally accessible socialized healthcare and public elementary education (Misión Permanente de Costa Rica ante Naciones Unidas 2024).
But what of the inexplicably high prevalence of disability, where so many have deemed development a success? According to the National Survey on Disability, this disability prevalence is even higher among Costa Rican women (60.9%) versus Costa Rican men (39.1%) (INEC 2019).
In developing countries, disability prevalence is generally exacerbated by poverty-related risks like malnutrition, limited healthcare access, unsafe work conditions, environmental pollution, and inadequate water and sanitation facilities (World Bank 2024).
Such disparities in healthcare access, water and sanitation, ecology and other indicators of progress underscore the United Nations’ call for disability-disaggregated data to pinpoint and address structural inequalities (United Nations 2019; United Nations Development Group 2011; United Nations Development Programme 2021).
In this way, disability prevalence can potentially not only reflect a range of social and economic inequalities but may also illuminate the reciprocal impact of systemic challenges on mulitply marginalized people.
What remains as a question with Costa Rica, then, is what is going on? Why is there such high disability prevalence, especially among women, when key indicators seem to suggest this should not be the case here?
This paradoxical situation of disabled development also raises important questions about both the nature of subjective well-being and its measurement. In this first study, I hope to contribute to theory and to the demographic study of evaluative well-being by investigating this paradox.
Disability in the 2030 Agenda
The 2030 Agenda for Sustainable Development with its 17 sustainable development goals (SDGs) and 169 targets sets out to try to end poverty and the unequal access to resources that produces it (United Nations 2016). The overarching stated value of the 2030 Agenda is to “leave no one behind” (United Nations 2020b) because development would not be sustainable if only some subpopulations were a part of it.
The words “disability” or “persons with disabilities” can be found 11 times in the 2030 Agenda, and “persons in vulnerable situations” appears six additional times. To leave no one behind requires first disaggregating data, and by more than just gender, age, race/ethnicity and location, in order to unmask inequalities so that they can be addressed (United Nations 2020a).
The UN Convention on the Rights of Persons with Disabilities (CRPD) and its Optional Protocol (OP) were adopted in 2006 with the objective of extending to disabled people equal human rights and respect (United Nations 2006). This provides the international backdrop of disability justice in development.
A recent study using DHS and MICS data, along with the Multidimensional Poverty Index, considered questions about the magnitude of deprivations and poverty for disabled people in consideration of sustainable development (Pinilla-Roncancio and Alkire 2021). Their main finding was a development-disability gap, which was greater in middle-income countries than in low-income countries.
As of the 13th Session of the CRPD, 190 States have ratified the Convention (United Nations Office at Geneva 2024). The CRPD recently reviewed Costa Rica’s progress on disability rights. They praised its employment initiatives and accessible tourism. Experts inquired about efforts to ensure outdoor activities are accessible and raised concerns about abortion policies, especially in cases involving intellectual disabilities and violence. Costa Rica stressed its commitment to disabled people’s access to public and private services and equal opportunity, and the report overall was a positive one (United Nations Office at Geneva 2024b).
Social Stress Process Model
Sociologists can understand the relationship between social factors and social outcomes (such as life evaluation or evaluative wellbeing) through the social stress process (Pearlin 1989). A fundamental principle of the stress process model is that well-being outcomes are not randomly distributed in society. When systematic gaps are observed, they are thought to reflect social, structural conditions that systematically disadvantage subgroups (Pearlin 1999).
It’s important to not discount the power of that which moderates the relationships between those elements of social structure (i.e. “the constellation of stressors”) and the outcomes of interest (i.e. wellbeing). In other words, just because multiply marginalized groups are structurally disadvantaged, they are not necessarily going to score lower on outcomes that reflect evaluative wellbeing and similar metrics.
That sort of paradox or surprise too can be accounted for via the social stress process, because of the presence of possible coping resources, themselves an elaborate constellation. Researchers found such an explanation to a paradox they initially observed in mental wellbeing data for Black Americans (Louie et al. 2022).
The paradox they had observed was that even though the group (Black Americans) experience greater burdens, which would suggest lower mental health scores, the group also possesses more coping resources, such as self-esteem and social support, which partially mediated the influence of racism on their mental health. It doesn’t mean the racism wasn’t bad, just that the coping resources were magnificent.
Social stress model remains the conceptual base for the sociological study of how social factors and evaluative well-being are related today (George 2014:251). It has also been the basis of considerable scholarship on related sociological topics, such as understanding links between education and health (Ross and Wu 1995), or minority stress and mental health in gay men (Meyer 1995).
We do these kinds of analyses because subpopulations can have categorically different levels of an important outcome, and the explanation may be found in the left side of the equation, within the constellation of stressors which reflect the social structure itself.
These explanations may also be found in the array of coping resources that mediates or moderates the effects. This framing offers a sociological framework for identifying inequalities that moves beyond describing differences in frequency distributions.
Moving beyond describing differences in life well-being across subnational regions, an analysis grounded in the social stress model will seek to identify how the subgroups meaningfully differ. By applying the social stress model, researchers may also identify underlying structural issues potentially influencing these wellbeing gaps.
Capability Approach
Predominant frameworks for measuring well-being today involve standard indicators such as income or consumption. Income serves as an initial proxy for the standard of living and, where it may fall short, researchers often supplement it with measures of consumption, according to non-capability researchers using traditional, resource-based measures of wellbeing (Burchardt and Hick 2018:43).
The resource-based models also include additional non-monetary indicators, such as assets and access to basic needs like housing, water, education, and healthcare. The OECD Wellbeing Framework, shared in the Methods discussion: About the Cantril ladder, is mostly a resource-based model for measuring multidimensional well-being.
In contrast to conceiving of wellbeing in terms of resources, the capability approach, developed by Amartya Sen (Sen 1987; Sen 1999), and elaborated by Martha Nussbaum (2000, 2011) is a normative framework that emphasizes the importance of individuals having the opportunities and freedoms to achieve well-being in ways that they value. This approach challenges traditional and most modern metrics by focusing on what people are actually able to do and to be, rather than solely on what they have.
A central tenet of the capability approach is the enhancement of possibilities, looking beyond economic success to a broader range of individual freedoms and opportunities. Nussbaum further contributes to this framework and notes that protecting these capabilities should be a priority in justice considerations (Nussbaum 2011).
The capability approach is useful for assessing inequalities in wellbeing across groups. It takes into account a wide range of factors that affect individuals’ real freedoms and choices. The capability approach’s view of wellbeing makes it a suitable theoretical lens through which to consider the relationships between education, healthcare, disability, discrimination and life evaluation among subpopulations of women.
For additional methods discussion, including the limitations of the evaluative wellbeing measures from a capability approach perspective, and the concern for adaptive preferences, please see the Methods discussion: About the Cantril ladder.
Hypothesis
Ethnicity (binary indicator: White/Not White) and subnational region (7 provinces in Costa Rica) represent structural inequalities associated with the wellbeing of disabled women, measured as life evaluation by the Cantril ladder. Specifically, disabled women of color living in disadvantaged regions will have lower subjective wellbeing compared to their non-disabled neighbors and to other disabled women in more highly resourced regions.
Contributions
This study fills a critical data gap in country, adding to the 2018 National Survey on Disability data and underscoring the importance of disability-disaggregated subjective wellbeing data. It also investigates a paradoxical empirical situation of high evaluative wellbeing coupled with high disability prevelance, both in aggregate. Given how important it is for disabled women to not be left behind, this situation poses important questions about the nature of wellbeing.
This study also includes a possible empirical test of an aspect of the capability approach, by exploring the concept of adaptive preferences as applied to the Cantril ladder’s measurement of well-being among disabled women in Costa Rica.
By examining the extent to which disabled women report unexpectedly high evaluative wellbeing, this study could affirm the Cantril ladder’s ability to capture aspects of wellbeing that are not influenced by such adaptation.
If the study does find that disabled women in Costa Rica report high evaluative wellbeing, such a finding would suggest that the Cantril ladder may be capturing the influence of adaptive preferences on well-being assessments.
From a theoretical perspective, this would suggest further investigation into the conditions under which adaptive preferences operate and their limits in contributing to the Cantril ladder measure of subjective wellbeing or life evaluation.
References
Methods
Data Source
This research will analyze the IPUMS MICS Round 6 for Costa Rica. Please find a Methods discussion: About IPUMS MICS on the home page for this prospectus.
Dependent Variable: Life Evaluation Measure
The outcome measure of evaluative well-being in this study is the Cantril ladder item, from IPUMS MICS Round 6. See also: Methods discussion: About the Cantril ladder.
Key Independent Variable: Constructed Disability Indicator
Disability is functional impairment in this study. The indicator of disabled here is constructed from six IPUMS MICS ordinal survey items that ask respondents to report their level of functional impairment.
Women were asked whether they had difficulty: seeing, hearing, walking, remembering and concentrating, communicating, and with self care. For each item, women could report that they have ‘no difficulty’, ‘some difficulty’, ‘a lot of difficulty’, or whether they cannot perform the given function at all.
If women report at least ‘some difficulty’ in at least one domain, they are coded as disabled in the present study. While this leads to higher prevalence rates than studies which focus on so-called ‘severe disability’, my intention here is to include as many disabled women as possible. I had also hoped to use a cut-off that would result in overall disability prevalence rates consistent with the National Survey on Disability carried out during the same year in Costa Rica (INEC 2019), and this threshold accomplishes that comparability.
Other Independent Variables
ethnicity: constructed binary indicator; Costa Rica is about 2/3 White and due to data limitations, ethnicity of household head needed to be binned into White and People of Color (Non-White respondents).
marital: binary indicator constructed using marital status; categories were combined so that women were 1=married or partnered, and 0=not married, with the latter including women who reported being divorced, widowed, separated or never married.
education: highest level of school attended by the woman, categorized into four ordered levels (less than primary, primary, secondary, or tertiary/higher/university).
healthinsur: binary indicator for whether a woman has health insurance.
region: reports the categorical, geographic subnational region according to which of the seven provinces the woman lives within in Costa Rica.
Generalized Ordered Logit Model
Let Yi represent the ordered response for life evaluation (lsbin), for individual i. The generalized ordered logistic regression model for estimating the probabilities of Yi ≤ j falling into specific categories given the predictors is:
logit[P(Yi ≤ j)] = αj -
(β1educationi +
β2healthinsuri +
β3disabledi +
β4regioni +
β5ethnicityi +
β6lnagei +
β7wealthi)
Where:
- j indexes the cutpoints (or thresholds) for each category of lsbin, which is Cantril ladder with 0-4 binned, excluding the highest category.
- αj are the intercept terms or threshold parameters for each category j.
- β1 through β7 are the coefficients associated with the predictors.
This model was selected because of the nature of the dependent variable, where categories on the Cantril ladder are ordered (levels on the ladder) but the distances between rungs are not meaningful. While the distribution’s shape is not specifically constrained, the proportional odds assumption must be maintained.
This implies that the effect of model factors, such as disability, remains constant across the scale of the outcome. Therefore, if being disabled influences the likelihood of a person rating their well-being as a 9 on the 0-10 Cantril ladder, for example, it should similarly affect their ratings at 8 or 5, or in the 0-4 low bin, with all other factors held constant.
I use the user-written Stata command gologit2 to test the proportional odds assumptions for each term. This command fits a partial proportional odds model—unlike ologit, which assumes full proportional odds—allowing for varying effects of some predictors across the Cantril ladder’s outcome levels.
gologit2 computes Wald tests to check the consistency of predictor effects across categories. If the effects are not significantly different, the proportional odds assumption is supported for those predictors. If the assumption holds only for certain predictors, gologit2 can adjust to fit a partial proportional odds model accordingly.
Preliminary Results
Figure 1: Life Evaluation by Province and Disability Status
Note. This graph shows kernel density plots for the life evaluation outcome measure (binned Cantril ladder) for Heredia, the wealthiest province, and for Limón, the least wealthy province. Noteworthy is the well-being penalty associated with disability for women living in Limón. Despite having the fewest economic resources, non-disabled women in Limón have the highest life satisfaction of women in Costa Rica.
Table 1: Regional Disability Prevalence by Country
| Disability Indicator |
Costa Rica | Cuba | Dominican Republic |
Honduras | Suriname | Total |
|---|---|---|---|---|---|---|
| Not Disabled | 4,554 51.10% |
7,334 80.51% |
16,426 68.49% |
12,817 57.36% |
6,033 70.70% |
47,164 67.84% |
| Disabled | 3,663 48.90% |
1,608 19.49% |
6,562 31.51% |
8,024 42.64% |
2,500 29.30% |
22,357 32.16% |
| Total | 8,217 100.00% |
8,942 100.00% |
22,988 100.00% |
20,841 100.00% |
8,533 100.00% |
69,521 100.00% |
Note. These are IPUMS MICS Round 6 data (2018-2019) for countries in Latin America and Caribbean (LAC) Region. This is a frequency table of the disability indicator used in this study, in which a person is disabled if they have at least ‘some’ difficulty in at least one functional domain. This coding results in a prevalence rate closer to the 2018 National Survey on Disability conducted by INEC-Costa Rica, which found 50.7% of women to be mild-to-moderately disabled.
Table 2: Regional Severe Disability Prevalence by Country
| Disability Indicator |
Costa Rica | Cuba | Dominican Republic |
Honduras | Suriname | Total |
|---|---|---|---|---|---|---|
| Not Disabled | 7,470 90.58% |
8,807 98.50% |
22,177 95.62% |
19,256 91.73% |
8,221 96.34% |
65,931 94.84% |
| Disabled | 747 9.42% |
135 1.50% |
811 4.38% |
1,585 8.27% |
312 3.66% |
3,590 5.16% |
| Total | 8,217 100.00% |
8,942 100.00% |
22,988 100.00% |
20,841 100.00% |
8,533 100.00% |
69,521 100.00% |
Note. These are based on IPUMS MICS Round 6 data (2018-2019). This is a frequency table of an indicator of severe disability, in which a person is disabled if they have at least ‘a lot’ of difficulty in at least one functional domain.
Figure 2: Life Evaluation for Women in IPUMS MICS 6 Costa Rica, 2018 (n = 7483)
Note. This graph shows the distribution of lsladder for all surveyed women in Costa Rica with levels 0-4 binned together. The Cantril ladder outcome measure is initially an 11-level ordinal variable but, as is shown in this graph, only 3% of women rated their life evaluation 0-4 on that scale. And while I believe it is very important to tell their stories as they are the women who are most left behind, it’s mathematically unhelpful to not bin those in the low categories together. This graph is present to show why the outcome measure is binned and transformed from an 11-level variable into a 7-level variable.
Table 3: A Few Findings from the First Generalized Ordered Logit Estimate, for discussion (n = 6971)
| Variable/Level | Direction | Coefficient | P>|t| | 95% CI |
|---|---|---|---|---|
| Education (Tertiary/Higher) | Positive | 1.385 (Level 4) | 0.013 | [0.299, 2.470] |
| Health Insurance (Yes) | Positive | 0.255 (Levels 4, 5, 6) | 0.023 | [0.035, 0.475] |
| Disabled | Negative | -0.692 (Level 4) | 0.010 | [-1.220, -0.164] |
| Disabled | Negative | -0.826 (Level 5) | 0.000 | [-1.095, -0.557] |
Note. Tertiary or higher education, and having health insurance, are positively associated with higher life evaluation levels, supporting Hypothesis 1a. Conversely, and as expected, being disabled is significantly negatively associated with life evaluation, which partially aligns with Hypothesis 2 but also suggests that while disability might involve adaptive preferences, it still poses challenges to subjective well-being. I don’t think it’s the disabled women who are buoying the Costa Rican subjective well-being targets. I think they are being left behind, and this, along with Figure 1, is early affirmation, or a focal point for the additional data discoveries to be revealed in the remaining analyses from Study 1. This table is just a few select findings from the ordered logit estimate.
Limitations
While the popularity of well-being economy and subjective well-being measures is going strong, limitations span from methodological challenges to the subjective nature of an outcome measure such as the Cantril ladder. Limitations of the outcome measure are discussed in the Methods discussion: Cantril ladder section.
While exploring adaptive preferences is an intentional part of this study, it also brings a limitation. Adaptive preferences can be complex to measure and to interpret because they involve subjective adjustments to aspirations based on circumstance, which might not be accurately assessed through survey data. There is always a risk of response bias/social desirability also.
And even while a life course perspective is likely to aid in understanding observed patterns, and aid in interpreting results, the same effects possibly influencing people can include factors that are unexamined altogether. This could reflect omitted variable bias perhaps, or some other confound.
I’m mindful of the critique raised by Kimberlé Crenshaw, that unless an intersectional analysis is taken to a political level, all we’re doing is describing people differently. In the unlikely event that it is all I end up doing with this dissertation research, then I too will have done nothing.
I also feel that there are some limitations to using the Washington Group Short Set (WG-SS) disability items. They have evolved to be accepted as a sort of unequivocal gold standard for comparative survey data science with disability disaggregated data.
They tout themselves as the social model variables, yet they still lean heavily towards “personal problem” rather than “social issue” and, in my opinion, the risk here is that our research, even while doing what’s recommended, could devolve into “counting crips” like so much medical research does. As one weary, disabled researcher, I think this is something we should be talking about as well, and which I will take up to some extent in study 2.
Stata Code
Loading and Merging Data
/*
a quick HELLO to public website guests - my old Stata tutorials
are in dire need of replacement, and good news!
This summer that will happen!
The things you've emailed about most often will be covered,
and I will use this dissertation research to demo how-to-do
lots of improved graphics in Stata 18.0 - it will be fun - stay tuned!
*/
clear all
*This is an IPUMS MICS data extract, courtesy of ipums.org
/* IPUMS MICS women's sample weight is weightwm;
it is the inverse of the probability of inclusion
- psu = cluster in this survey
- centered is a choice re: handling single PSU in stratum;
*/
svyset [pweight=weightwm], ///
psu(cluster) strata(stratum) singleunit(centered)
/*
Study population and sampling designs:
The sampling units were women aged 15-49.
MICS uses a stratified, two-stage clustering design.
Other surveys that were used for validity checks:
The DHS program uses a two-stage stratified sampling design.
GWP uses a stratified random sampling design.
*/
/***************************
MERGING
***************************/
* women's sample and HH sample are separate, and ethnicity is only available
* for head of household. So I download HH sample and merge them many-to-one
* Open the women data file
clear all
use "disswm.dta" , clear
* Sort cases by ID variables
sort cluster hhno linewm
* Save the women file.
save "disswm-sort.dta", replace
* Open the household file
clear all
use "disshh.dta" , clear
* Sort cases by ID variables
sort cluster hhno
* Save the household file
save "disshh-sort.dta", replace
* re-open the women's sorted file
clear all
use "disswm-sort.dta", clear
* Merge the household sorted file onto the women sorted file by many to one
merge m:1 sample cluster hhno ///
using "disshh-sort.dta", ///
keepusing (ethnic* stratum imicsstratum weighthh religionhhead numwm numch)
* Save the MERGED women's file. It's now the women's sample I wanted, with the
* additional, household level variables attached - cool
save "dissmerged.dta", replace
Cleaning Data and Creating Indicators
*to set cluster as psu and svyset data:
svyset [pweight=weightwm], ///
psu(cluster) strata(stratum) singleunit(centered)
/***************************
CLEANING
***************************/
clear all
use "dissmerged.dta" , clear
*CLEANING - recoding NR, NIU as .
recode resultwm glasses hearaid healthinsur lit ///
diff* discrim* safehome safewalk lslastyr lsnextyr (7/9=.)
recode langrespond (9999=.)
recode marst agewm edlevelwm lsladder (98/99=.)
recode geo1_cr (188998=.) //missing due to incomplete questionnaire
*lsbin
gen lsbin = .
replace lsbin = lsladder
recode lsbin (0/4=4)
label define lsbincr 4 "lsladderlow" 5 "lsladder5" 6 "lsladder6" ///
7 "lsladder7" 8 "lsladder8" 9 "lsladder9" 10 "lsladder10", replace
label values lsbin lsbin
tab lsladder lsbin //inspect
* lshurdle (for later, I have an idea for a hurdle model study)
gen lshurdle = (lsladder > 4) if lsladder <= 10
label variable lshurdle "lsladder binary outcome"
label define lshurdle_lbl 0 "lsladder 0-4" 1 "lsladder 5-10"
label values lshurdle lshurdle_lbl
tabulate lsladder lshurdle, missing
tab lshurdle //I think investigating a hurdle model makes sense empirically
* there are social factors women have, and don't have, that determine whether
* they surpass this hurdle. In Costa Rica, most women DO surpass. But I am
* interested in who does not get over the hurdle. Their stories are so valuable.
* side project I guess, or an implication for further study after this one,
* but I won't leave these women behind
*CREATING INDICATORS
*create indicator for lowladder
gen lowladder = (lsladder >= 0 & lsladder <= 4)
label variable lowladder "Indicator for lsladder 0-4"
// it's the same as lshurdle I just want it separate
* ethnic
gen ethnic = .
lab var ethnic "Ethnicity"
replace ethnic = ethnic_cr
recode ethnic (3=6) (4=3) (5=4) (6=5) (7=6) (90=6)
label define ethnic 1 "Indigenous" 2 "Black/Afro-CR" 3 "Mestizo" ///
4 "Mulatto" 5 "White" 6 "Other/none" , replace
recode ethnic (98/99=.)
label values ethnic ethnic
tab ethnic ethnic_cr //inspect
* ethnicity - this is a Whiteness indicator
gen ethnicity = ethnic
recode ethnicity (1/4=0) (5=1) (6=0)
lab var ethnicity "Ethnicity"
label define ethnicity 0 "Black/POC" 1 "White"
label values ethnicity ethnicity
tab ethnic ethnicity //inspect
* discrimination
local discrimvars ///
"discrimgender discrimsexorien discrimage discrimrelig discrimdis discrimother"
gen discrimination = 0
foreach var of local discrimvars {
replace discrimination = 1 if `var' == 1
}
label variable discrimination "Discrimination Indicator"
label define discrimination 0 "Not Discriminated" 1 "Discriminated"
label values discrimination discrimination
tab discrimination //inspect
* wealth
gen wealth = (wscore > 0)
label variable wealth "wscore indicator"
label define wealth_lbl 0 "negative wscore" 1 "positive wscore"
label values wealth wealth_lbl
tab wealth //inspect
* marital
gen marital = marst
lab var marital "married indicator"
recode marital (10=1) (20/30=0)
lab def marital 0 "not partnered" 1 "married/in union"
lab val marital marital
tab marital marst //inspect
* education
rename edlevelwm education
recode education (10=1) (20=2) (41=3) (42=4)
lab def education 1 "less than primary" 2 "primary" 3 "secondary" ///
4 "tertiary/higher/uni"
lab val education education
tab education //inspect
* safety (for study 3)
*safewalk
*safehome
* lnage
gen lnage = ln(agewm)
lab var lnage "ln-transformed age"
* indicators for each region in CR
gen CR1 = (geo1_cr == 1)
label variable CR1 "San José Indicator"
label define cr1_labels 0 "No" 1 "Yes"
label values CR1 cr1_labels
tab CR1 //verify
gen CR2 = (geo1_cr == 2)
label variable CR2 "Alajuela Indicator"
label define cr2_labels 0 "No" 1 "Yes"
label values CR2 cr2_labels
tab CR2 //verify
gen CR3 = (geo1_cr == 3)
label variable CR3 "Cartago Indicator"
label define cr3_labels 0 "No" 1 "Yes"
label values CR3 cr3_labels
tab CR3 //verify
gen CR4 = (geo1_cr == 4)
label variable CR4 "Heredia Indicator"
label define cr4_labels 0 "No" 1 "Yes"
label values CR4 cr4_labels
tab CR4 //verify
gen CR5 = (geo1_cr == 5)
label variable CR5 "Guanacaste Indicator"
label define cr5_labels 0 "No" 1 "Yes"
label values CR5 cr5_labels
tab CR5 //verify
gen CR6 = (geo1_cr == 6)
label variable CR6 "Puntarenas Indicator"
label define cr6_labels 0 "No" 1 "Yes"
label values CR6 cr6_labels
tab CR6 //verify
gen CR7 = (geo1_cr == 7)
label variable CR7 "Limón Indicator"
label define cr7_labels 0 "No" 1 "Yes"
label values CR7 cr7_labels
tab CR7 //verify
* lshappy
generate lshappyorig = lshappy
recode lshappy (5=1) (4=2) (3=3) (2=4) (1=5) (8/9=.)
label define lshappy 1 "very unhappy" 2 "a little unhappy" ///
3 "neither unhappy nor happy" 4 "somewhat happy" 5 "very happy"
label values lshappy lshappy
tab lshappyorig lshappy //inspect
* region
rename geo1_cr region
recode region ///
(188001=1) (188002=2) (188003=3) (188004=4) ///
(188005=5) (188006=6) (188007=7)
label define region 1 "San José" 2 "Alajuela" 3 "Cartago" 4 "Heredia" ///
5 "Guanacaste" 6 "Puntarenas" 7 "Limón"
label values region region
tab region //inspect
* disabled
local diffvars "diffsee diffhear diffwalk diffremcon diffcare diffcom"
gen disabled = 0
foreach var of local diffvars {
replace disabled = 1 if inlist(`var', 2, 3, 4)
}
label variable disabled "Disability Indicator"
label define disabled 0 "Not Disabled" 1 "Disabled"
label values disabled disabled
tab disabled
* disabled_lot
gen disabled_lot = 0
foreach var of local diffvars {
replace disabled_lot = 1 if inlist(`var', 3, 4)
}
label variable disabled_lot "A Lot Disability Indicator"
label define disabled_lot 0 "Not Disabled" 1 "A lot Disabled"
label values disabled_lot disabled_lot
tab disabled_lot
/* quick explainer:
How to efficiently create a binary indicator based on the presence or absence
of 'a lot' of functional disability across six domains that are ordinal measures
This loop iterates over the list of disability variables, setting
the new 'disabled' variable to 1 if any of the disability variables are
3 or 4, which represents at least 'a lot' for level of difficulty in that
functional domain, thus moving the disability Indicator of 'disabled' to 1.
Else, 'disabled' stays at zero.
This construction is consistent with the Washington Group recommendations for
disability disaggregated data, which is the basis for the MICS and other
survey tools that include these items, and is what other researchers
follow when using these items in their research
*/
********** Save the merged + cleaned CR women's file ************
save "disscleaned.dta", replace
/***************************************************
Mario Flag Save Checkpoint
***************************************************/
Descriptives
/**************************************************
Let's-a-go!
_
_( }
-= _ << \
`.\__/`/\\
-= '--'\\ `
-= //
\)
**************************************************/
clear all
use "disscleaned.dta" , clear
svyset [pweight=weightwm], ///
psu(cluster) strata(stratum) singleunit(centered)
* lsladder overall in CR
svy, subpop(CR): tab lsbin //see bar graph
/*
Explaining: Even though 'weightwm' is a probability weight,
typically used to correct for sample selection biases, and
ensure representativeness of estimates, I've labeled it as
an a-weight in this kernel density plot. This is okay here
because my purpose is to adjust the contribution of each
observation to reflect the overall shape of the distribution
accurately. Concerns about more precise variance estimation,
which might be +/- this, don't so much matter for this
purely visual descriptive.
*/
* Study 1 Figure 1
twoway (kdensity lsbin if country==188 & region==4 & disabled==1 ///
[aweight=weightwm], kernel(epanechnikov) lwidth(thick) lcolor(orange) ///
lpattern(solid) legend(label(1 "Heredia - Dis."))) ///
(kdensity lsbin if country==188 & region==4 & disabled==0 ///
[aweight=weightwm], kernel(epanechnikov) lwidth(thick) lcolor(orange) ///
lpattern(dash) legend(label(2 "Heredia - Nond."))) ///
(kdensity lsbin if country==188 & region==7 & disabled==1 ///
[aweight=weightwm], kernel(epanechnikov) lwidth(thick) lcolor(navy) ///
lpattern(solid) legend(label(3 "Limón - Dis."))) ///
(kdensity lsbin if country==188 & region==7 & disabled==0 ///
[aweight=weightwm], kernel(epanechnikov) lwidth(thick) lcolor(navy) ///
lpattern(dash) legend(label(4 "Limón - Nond."))) ///
, title("Life Evaluation by Province in Costa Rica - Comparisons") ///
name(CR_ladders_comp, replace) ///
yscale(range(0 .25)) ytick(0(.05).25) xsca(range(4 10)) ///
xlabel(4 "ladder 0-4" 6 "ladder 6" 8 "ladder 8" 10 "ladder 10", valuelabel) ///
xtitle("Cantril Ladder")
/* This graph shows the kernel density plots for the life evaluation
outcome measure (binned Cantril ladder) for Heredia, the wealthiest
province, and Limón, the least wealthy. What strikes me is the disability
penalty felt by women in the lowest resource province.
*/
*Table 1
//disabled - weighted frequencies and CIs
* Heredia - White
svy, subpop(if ethnicity == 0 & region == 4): tabulate disabled, ci
* Heredia - POC
svy, subpop(if ethnicity == 1 & region == 4): tabulate disabled, ci
* Limón - White
svy, subpop(if ethnicity == 0 & region == 7): tabulate disabled, ci
* Limón - POC
svy, subpop(if ethnicity == 1 & region == 7): tabulate disabled, ci
Analysis
/******************************************************
ANALYSIS
******************************************************/
clear all
use "disscleaned.dta" , clear
svyset [pweight=weightwm], ///
psu(cluster) strata(stratum) singleunit(centered)
gsvy: gologit2 lsbin i.education i.healthinsur i.disabled i.region ///
i.ethnicity c.lnage i.wealth, autofit(.01)
eststo study1hyp1a
// I realize the following is not very readable. I put a tiny table into
// the end of my Preliminary Results section, with a note. I'm including this
// here in case anyone wants to glance at it. Thanks.
/*
Testing parallel lines assumption using the .01 level of significance...
Step 1: Constraints for parallel lines imposed for 4.region (P Value = 0.9397)
Step 2: Constraints for parallel lines imposed for 1.ethnicity (P Value = 0.4635)
Step 3: Constraints for parallel lines imposed for lnage (P Value = 0.4524)
Step 4: Constraints for parallel lines imposed for 3.region (P Value = 0.4424)
Step 5: Constraints for parallel lines imposed for 7.region (P Value = 0.2939)
Step 6: Constraints for parallel lines imposed for 5.region (P Value = 0.2536)
Step 7: Constraints for parallel lines imposed for 1.healthinsur (P Value = 0.2729
> )
Step 8: Constraints for parallel lines imposed for 6.region (P Value = 0.1956)
Step 9: Constraints for parallel lines imposed for 2.region (P Value = 0.0919)
Step 10: Constraints for parallel lines are not imposed for
2.education (P Value = 0.00247)
3.education (P Value = 0.00000)
4.education (P Value = 0.00000)
1.disabled (P Value = 0.00033)
1.wealth (P Value = 0.00203)
Wald test of parallel lines assumption for the final model:
Adjusted Wald test
F( 45, 423) = 1.21
Prob > F = 0.1718
An insignificant test statistic indicates that the final model
does not violate the proportional odds/ parallel lines assumption
Survey: Generalized Ordered Logit Estimates
Number of strata = 32 Number of obs = 6,971
Number of PSUs = 499 Population size = 1,243,968
Design df = 467 F(39, 429) = 12.00
Prob > F = 0.0000
-----------------------------------------------------------------------------------
| Linearized
lsbin | Coefficient std. err. t P>|t| [95% conf. interval]
------------------+----------------------------------------------------------------
4 |
education |
primary | .3569874 .4693462 0.76 0.447 -.5653045 1.279279
secondary | .8468945 .4694965 1.80 0.072 -.0756929 1.769482
tertiary/highe~i | 1.385164 .5524942 2.51 0.013 .2994816 2.470847
|
healthinsur |
Yes | .2551915 .1120413 2.28 0.023 .0350239 .475359
|
disabled |
Disabled | -.6920357 .2689162 -2.57 0.010 -1.220471 -.1636001
|
region |
Alajuela | .1385226 .1158187 1.20 0.232 -.0890677 .3661128
Cartago | .1246641 .1246942 1.00 0.318 -.1203671 .3696953
Heredia | .2312699 .1935252 1.20 0.233 -.1490182 .611558
Guanacaste | .3783973 .1310193 2.89 0.004 .1209369 .6358576
Puntarenas | .2956824 .1296342 2.28 0.023 .0409438 .550421
Limón | .2782322 .1189915 2.34 0.020 .0444071 .5120573
|
ethnicity |
White | .0989506 .0730357 1.35 0.176 -.0445686 .2424698
|
lnage | .1033463 .1273937 0.81 0.418 -.1469895 .3536822
|
wealth |
positive wscore | .8542576 .1952397 4.38 0.000 .4706005 1.237915
|
_cons | 1.711685 .7581226 2.26 0.024 .221931 3.201439
------------------+----------------------------------------------------------------
5 |
education |
primary | .0842846 .3941993 0.21 0.831 -.6903394 .8589087
secondary | .4272154 .3930549 1.09 0.278 -.3451598 1.199591
tertiary/highe~i | .8316412 .4371863 1.90 0.058 -.0274546 1.690737
|
healthinsur |
Yes | .2551915 .1120413 2.28 0.023 .0350239 .475359
|
disabled |
Disabled | -.8259423 .1367817 -6.04 0.000 -1.094726 -.5571585
|
region |
Alajuela | .1385226 .1158187 1.20 0.232 -.0890677 .3661128
Cartago | .1246641 .1246942 1.00 0.318 -.1203671 .3696953
Heredia | .2312699 .1935252 1.20 0.233 -.1490182 .611558
Guanacaste | .3783973 .1310193 2.89 0.004 .1209369 .6358576
Puntarenas | .2956824 .1296342 2.28 0.023 .0409438 .550421
Limón | .2782322 .1189915 2.34 0.020 .0444071 .5120573
|
ethnicity |
White | .0989506 .0730357 1.35 0.176 -.0445686 .2424698
|
lnage | .1033463 .1273937 0.81 0.418 -.1469895 .3536822
|
wealth |
positive wscore | .6032923 .1296251 4.65 0.000 .3485716 .858013
|
_cons | 1.047825 .6380082 1.64 0.101 -.2058972 2.301548
------------------+----------------------------------------------------------------
6 |
education |
primary | -.1056894 .3810996 -0.28 0.782 -.8545717 .643193
secondary | -.0126009 .3775782 -0.03 0.973 -.7545635 .7293617
tertiary/highe~i | .5073833 .394781 1.29 0.199 -.2683838 1.28315
|
healthinsur |
Yes | .2551915 .1120413 2.28 0.023 .0350239 .475359
|
disabled |
Disabled | -.6788797 .1037618 -6.54 0.000 -.8827775 -.4749819
|
region |
Alajuela | .1385226 .1158187 1.20 0.232 -.0890677 .3661128
Cartago | .1246641 .1246942 1.00 0.318 -.1203671 .3696953
Heredia | .2312699 .1935252 1.20 0.233 -.1490182 .611558
Guanacaste | .3783973 .1310193 2.89 0.004 .1209369 .6358576
Puntarenas | .2956824 .1296342 2.28 0.023 .0409438 .550421
Limón | .2782322 .1189915 2.34 0.020 .0444071 .5120573
|
ethnicity |
White | .0989506 .0730357 1.35 0.176 -.0445686 .2424698
|
lnage | .1033463 .1273937 0.81 0.418 -.1469895 .3536822
|
wealth |
positive wscore | .4757863 .1074383 4.43 0.000 .2646639 .6869087
|
_cons | .856723 .6247541 1.37 0.171 -.3709543 2.0844
------------------+----------------------------------------------------------------
7 |
education |
primary | -.3758635 .335906 -1.12 0.264 -1.035938 .2842107
secondary | -.3717216 .3400176 -1.09 0.275 -1.039876 .2964324
tertiary/highe~i | -.1536824 .3526895 -0.44 0.663 -.8467373 .5393725
|
healthinsur |
Yes | .2551915 .1120413 2.28 0.023 .0350239 .475359
|
disabled |
Disabled | -.505248 .0845964 -5.97 0.000 -.6714847 -.3390114
|
region |
Alajuela | .1385226 .1158187 1.20 0.232 -.0890677 .3661128
Cartago | .1246641 .1246942 1.00 0.318 -.1203671 .3696953
Heredia | .2312699 .1935252 1.20 0.233 -.1490182 .611558
Guanacaste | .3783973 .1310193 2.89 0.004 .1209369 .6358576
Puntarenas | .2956824 .1296342 2.28 0.023 .0409438 .550421
Limón | .2782322 .1189915 2.34 0.020 .0444071 .5120573
|
ethnicity |
White | .0989506 .0730357 1.35 0.176 -.0445686 .2424698
|
lnage | .1033463 .1273937 0.81 0.418 -.1469895 .3536822
|
wealth |
positive wscore | .4639559 .0905633 5.12 0.000 .2859939 .6419179
|
_cons | .4232443 .5871422 0.72 0.471 -.7305235 1.577012
------------------+----------------------------------------------------------------
8 |
education |
primary | -.2522958 .3201181 -0.79 0.431 -.881346 .3767544
secondary | -.4667982 .3362805 -1.39 0.166 -1.127608 .194012
tertiary/highe~i | -.548812 .3461701 -1.59 0.114 -1.229056 .1314319
|
healthinsur |
Yes | .2551915 .1120413 2.28 0.023 .0350239 .475359
|
disabled |
Disabled | -.3335258 .080704 -4.13 0.000 -.4921138 -.1749378
|
region |
Alajuela | .1385226 .1158187 1.20 0.232 -.0890677 .3661128
Cartago | .1246641 .1246942 1.00 0.318 -.1203671 .3696953
Heredia | .2312699 .1935252 1.20 0.233 -.1490182 .611558
Guanacaste | .3783973 .1310193 2.89 0.004 .1209369 .6358576
Puntarenas | .2956824 .1296342 2.28 0.023 .0409438 .550421
Limón | .2782322 .1189915 2.34 0.020 .0444071 .5120573
|
ethnicity |
White | .0989506 .0730357 1.35 0.176 -.0445686 .2424698
|
lnage | .1033463 .1273937 0.81 0.418 -.1469895 .3536822
|
wealth |
positive wscore | .245966 .090887 2.71 0.007 .0673679 .4245641
|
_cons | -.4915472 .5714029 -0.86 0.390 -1.614386 .6312919
------------------+----------------------------------------------------------------
9 |
education |
primary | -.5311873 .3267359 -1.63 0.105 -1.173242 .1108674
secondary | -.9089309 .3439856 -2.64 0.009 -1.584882 -.2329796
tertiary/highe~i | -1.337475 .3562704 -3.75 0.000 -2.037566 -.6373831
|
healthinsur |
Yes | .2551915 .1120413 2.28 0.023 .0350239 .475359
|
disabled |
Disabled | -.1888071 .0902699 -2.09 0.037 -.3661926 -.0114215
|
region |
Alajuela | .1385226 .1158187 1.20 0.232 -.0890677 .3661128
Cartago | .1246641 .1246942 1.00 0.318 -.1203671 .3696953
Heredia | .2312699 .1935252 1.20 0.233 -.1490182 .611558
Guanacaste | .3783973 .1310193 2.89 0.004 .1209369 .6358576
Puntarenas | .2956824 .1296342 2.28 0.023 .0409438 .550421
Limón | .2782322 .1189915 2.34 0.020 .0444071 .5120573
|
ethnicity |
White | .0989506 .0730357 1.35 0.176 -.0445686 .2424698
|
lnage | .1033463 .1273937 0.81 0.418 -.1469895 .3536822
|
wealth |
positive wscore | .145665 .1040206 1.40 0.162 -.0587414 .3500714
|
_cons | -.8669734 .580035 -1.49 0.136 -2.006775 .2728284
-----------------------------------------------------------------------------------