Multiple Regression with Discrete Dependent Variables (Pocket Guide to Social Work Research Methods)
Group Work Research is an essential resource for… More. Shelve Group Work Research. Historical Research by Elizabeth Ann Danto. What, exactly, was the Charity Organization Socie… More. Shelve Historical Research. When social workers draw on experience, theory, o… More. Key Concepts in Measurement by Brian E. Measurement refers generally to the process of as… More. Shelve Key Concepts in Measurement. Mixed Methods Research by Daphne Watkins. Finally, a practical guide to mixed methods resea… More.
Shelve Mixed Methods Research. Most social work researchers are familiar with li… More.
Narrative Inquiry by Kathleen Wells. Narrative inquiry allows for a detailed examinati… More. Shelve Narrative Inquiry. Needs Assessment by David Royse. In today's rapidly changing world, new social and… More. Shelve Needs Assessment. Participatory Action Research by Hal A.
As novel, complex social problems increase, espec… More. Shelve Participatory Action Research. Although practitioners do not often identify an e… More. Qualitative methods have become increasingly popu… More. Shelve Qualitative Methods for Practice Research. It is usually the case that ethical and pragmatic… More. Shelve Quasi-Experimental Research Designs. Randomized controlled trials RCTs are considere… More. Social work researchers often conduct research wi… More.
Secondary Data Analysis by Thomas P. In recent decades, social work and other social s… More. Shelve Secondary Data Analysis. Random sampling and random assignment are conside… More. Structural Equation Modeling by Natasha K. Structural Equation Modeling Sem has long been… More. Shelve Structural Equation Modeling. Survival Analysis by Shenyang Guo. Survival analysis is a class of statistical metho… More. Shelve Survival Analysis.
Qualitative synthesis within the family of system… More. Shelve Systematic Synthesis of Qualitative Research. In order to measure it, some researchers use the utilization of health care services as a proxy variable while others have proposed to evaluate access according to characteristics of the population such as insurance coverage or family income. Furthermore, some researchers use the characteristics of the delivery system as a measurement for access such as the distribution and availability of health care facilities [ 25 ].
In this study, definition of access was focused primarily on financial dimension, including health insurance. Other dimensions of access were not included in this analysis, partly due to data limitation. We approximate access by including two main components; the use of health services, and factors that facilitate or impede the use of health care services.
- Pocket Guides To Social Work Research Methods Series!
- Foopies Open Mic.
- Early Sprouts: Cultivating Healthy Food Choices in Young Children (NONE);
- Marketing international - 2e édition (Les Topos) (French Edition)!
We are aware that measuring utilization considers only one part of the population, i. Having access to health care services does not mean that one actually use medical care. This study focuses on health care utilization as a proxy measure of access, i. As such, we recognized the limitation of using utilization as a proxy measurement being incomplete to represent the concept of access. Egalitarians claim that an appropriate allocation of medical care according to need promotes health equality [ 26 ].
However, Culyer and Wagstaff [ 16 ] have shown that allocating health care expenditures according to need does not necessarily result in or promote equality of health. Instead, it will depend on the definition of need adopted. One of the most popular definitions establishes that need measures the care that is required to obtain the maximum possible health improvement within given resource constraints [ 27 ]. Therefore, unmet needs arise from the absence of care when resources were available.
In this study, the indicators used in the prediction of needed health care are demographic variables age-sex dummy variables plus health status and morbidity variables self-reported health problems and presence of chronic conditions. Both surveys have national coverage. CASEN is a biannual household survey that represents the national, regional, urban, and rural areas in Chile [ 28 ]. The survey has been carried out since ; however, we are limiting our analysis to surveys starting from because of changes in the CASEN questionnaires.
In this way we can maintain consistency in the variables that are analyzed. The main purpose of the survey is to describe the socioeconomic situation in Chile. It also includes a section on health status. The sampling method is multi-stage random sampling with geographical stratification and clustering. The first National Survey on Satisfaction and Out-of-Pocket Payment was carried out in , and there are no new versions of this survey to date [ 29 ]. It represents national and urban households. The sample design is strictly probabilistic, multistage 5 stages , geographical stratification and conglomerates.
The sample of the survey involves 4, households or 16, individuals. Using these two surveys, we assess equity in the use of health care services by two approaches. For the two-part model, we first estimated a logit model, where the dependent variable is a binary variable of whether the respondent used health care services.
Health care utilization data is known to have a skewed distribution with many of the people surveyed having no health care use in the recall period. Therefore, a logit model is more appropriate given the distribution of the data, which is highly concentrated in the tails. The results from the logit model will contribute to the second stage of the analysis. In the second part of the model we estimated a linear regression model for the frequency of health care use. Finally, we decomposed income-related inequalities in medical care use into contributions of need and non-need factors and estimated a horizontal inequity index using the Satisfaction and Out-of-Pocket Payment data.
The purpose of including the estimation of horizontal equity index analysis is to provide supplemental information for our equity analysis. While the two-part model estimation can provide information on whether and to what extent social-economic factors affect access to health care, estimated horizontal equity index can help put our two-part model analysis in a broad perspective of how equitable access to health care was distributed in the society as a whole. To better understand equity in health care use, it is imperative to identify factors associated with its use and assess the extent to which these factors may contribute to inequities in the system.
- Dillon and The Voice of Odin: 10th Anniversary Edition (The Dillon Adventures).
- On Likert scales, ordinal data and mean values - Achilleas Kostoulas!
- Alcohol-Based Handrub Improves Compliance With Hand Hygiene in Intensive Care Units?
- Whispers: Poetry from the Silence;
- Czasopismo zatwierdzone.
- Equity in health care utilization in Chile | SpringerLink.
- Randomized Controlled Trials | Phyllis Solomon Book | In-Stock - Buy Now | at Mighty Ape NZ?
- Feminism Confronts Technology.
- Research Methods in Psychology.
- Pocket Guides to Social Work Research Methods Series.
- The Hand Book of Spiritual and Physical Relationship Construction and Repair?
- 2nd Canadian Edition.
In order to identify such factors we estimated two-part models not only for examining health care utilization at an aggregate level but we also estimated separate models in the following categories of health care services: preventive care, general practitioner visits, specialty visits, and emergency care. To implement this analysis, we included the data collected by the CASEN survey for the years to We decided to use the two-part model for evaluating health care utilization because the decision on whether to use health care services and the quantity of use is based on a two-step process.
Generally, the initial visit to a physician or health service provider depends largely on the patient, while the following visits are associated with other factors such as the quality and satisfaction of the services, and the influence of the physician, among others. A measure of health care utilization like this, which includes the number of visits to physicians over a given period of time, is a discrete and non-negative value count. We have datasets with a large proportion of zeros, representing those who did not receive health care services during the recall period of data collection.
Zero counts and positive counts in health care utilization represent the actual level of use of medical services. A zero value does not represent a missing value, and they are required in order to understand the level of use of medical services [ 30 ]. Conceptually, the two-part model can solve the problem of excess of zeros and is a more appropriate model than using negative binomial or Poisson models [ 31 ].
According to Gerdtham [ 32 ], Polhlmeier and Ulrich [ 33 ] two-part models provide a better fit to health care utilization than negative binomial or Poisson models. We recognize that there is a debate on which model is more appropriate to handle the excess of zeros, either the two-part model or zero-inflated models, such as the zero-inflated Poisson and the zero-negative binomial.
More details on such debate can be found in Jones [ 34 ]. The dependent variable for the two-part model was the number of self-reported health care visits during a year. In order to estimate the model we constructed two variables. The first is a dichotomous variable indicating the use or non-use of health care services; the second variable indicates the number of health care visits. We repeat this model to analyze the other four types of health services under study, which are preventive visits, general practitioner visits, specialty visits, and emergency visits. The main explanatory variables for the CASEN dataset were geographical region, gender, age, marital status, availability of electricity, water and waste disposal, type of housing and housing ownership status, education level, insurance system, working hours, income, and health care payment.
We also used the Satisfaction and Out-of Pocket data to carry out a two-part model estimation. The analysis was conducted at the individual level since the two-part model is well suited to model individual level health care utilization data. This model allows us to address issues that cannot be addressed at the aggregate level like assessing separate effects of key variables on health care utilization among all individuals. The main explanatory variables were geographical region, gender, civil status, education level, work status, chronic disease, accidents, type of housing and housing ownership status, type of insurance, beneficiaries from the insurance system, dependent worker, emergency insurance, additional insurance, debt, work insurance, total number of people in the household, health care expenditures, income, health care satisfaction, and AUGE.
A widely used method to estimate equity is measuring it through the analysis of need factors, and whether factors other than need affect the utilization of health care. However, need is a concept that can be difficult to define and measure [ 27 ]. In this study we relied on demographics and health conditions as a proxy measured for need. We use the indirect method comparing the differences between actual need and need-standardized distributions for the probability of using health care during a year.
We specified a probit model with control variables to show the difference between need-predicted use and actual use [ 35 ]. Also, we computed need-standardized health care use with and without controls using ordinary least square OLS and probit models.
Looks like you do not have access to this content.
The functional form G can take different forms for probit, logit or other model. In this study we use a logit model. Using this method, we are able to measure the contribution of each factor and also the importance within the total contribution to inequality in health care use.
This method for decomposition holds only if we are working with linear regressions [ 37 ]. We need to use a nonlinear approximation in the case of nonlinear models, such as health care use when there is a large number of people who did not use health care services during the recall period. Residential status? Do you belong to an insurance system? Out-of-pocket health care payment: pay, sometimes pay, other type of payment, do not pay.
Health status? The response variable for the logit model is a dichotomous variable for health care use, referring to whether or not people have had any utilization of health care services during the period of analysis. The response variable for the OLS model is the frequency of health care use, excluding those who did not have any usage. In order to be consistent with the method of decomposing the concentration index, explanatory variables in the regression model are classified into three groups: income, need variables and other variables.
Need variables include gender, age, and health status. Other variables include geographical region, marital status, and availability of basic services such as electricity and water, type of housing, level of education, health insurance, health care payment, and AUGE coverage. Last level of study approved: no education, elementary school, high school, technical-professional school, technical training center, professional institute, university. Type of insurance system: Indigent card, Fonasa, Isapre, Capredena, Dipreca, other system, no insurance. Are you a dependent worker whose employer automatically deducts the payment for health insurance?
During the past year have you had to borrow money to pay health care costs? Overall health care satisfaction? Was your health problem covered by AUGE? The response variable for the OLS model is the frequency of health care use. We also organized the variables into three groups: income, need variables and other variables.
Need variables include age, gender and health status. In this survey health status is self-reported, which include chronic diseases and accidents. Other variables included in the model are civil status, region, education level, work status, type of housing, health insurance, health care satisfaction, and AUGE coverage. In addition, explanatory variables such as out-of-pocket payment and insurance may be endogenous. To test the endogeneity of some of the explanatory variables, we utilized a version of the Durbin-Wu-Hausman test using instrumental variables.
We concluded at the five per cent level that the null hypothesis that the variables are exogenous cannot be rejected. We also performed likelihood ratios to assess the fit of our model and omitted variables. Dependent variable for the logit model is a dichotomous indicator of whether a person has had health care during a particular year or not. Dependent variable for the OLS model is the number of health care use. Confidence intervals are presented in brackets [ ].
In the logit model, most of the variables were statistically significant at 5 per cent p-value, mainly due to the large sample size of 1,, individuals. However, in this study, we are more interested in evaluating practical significance; hence we want to estimate the size of the estimated coefficients and their impact on health care utilization. Also, we complement this study with individual models for the years to in order to assess which variables are important to predict use of health care services.
The results suggest that the strongest predictors of health care use are the years the survey was conducted, gender, marital status, availability of electricity and water, insurance, type of housing, schooling, natural logarithm of income, and AUGE. Our results also suggest that females are more likely to use health care services than males. The odds of using health care services for females were 1. Also, married people are more likely to use health care services than non-married, and our results suggest that the odds for married individuals are associated with Further, there is also a positive association for those individuals who have portable water and electricity in their homes to use more health care services than the individuals who do not have these basic services point estimate for water 0.
Glossary of Research Terms - SAGE Research Methods
Estimated results for the schooling variable are different from our expectation. We tend to expect that individuals with education will use more health care services than those individuals with no education, given the same health need. Our results, however, show that the odds of using health care services for those with education are only 0.
Likewise, being under the AUGE program increases the odds of health care use. It is worth noting that education and income usually result in higher use of health care services; however, educated persons tend to take a better care of their health and have fewer acute diseases Muller, [ 42 ]. This may explain the relationships we found. Although health need variables were included as a control in our model, however, it could be these variables were not adequately measuring health needs, hence education might capture that differences, thus this estimated results.
Among the 5 different OLS models we find that school, AUGE, and type of payment are strongest predictors for utilization of health care services, preventive care, practitioners visits, specialty visits, and emergency visits. When analyzing for preventive services, practitioner visits, specialty visits, and emergency visits, we find that people with education were associated with At the same time, individuals with education are more likely to use general practitioner visits 4.
Estimation result of two - part model for health care utilization for the Satisfaction and Out - of - Pocket Payment Survey. Dependent variable for the logit model is a dichotomous indicator of whether a person has had health care utilization during a particular year or not.
Dependent variable for the OLS model is frequency of health care use. For the logit model, estimated coefficients of the following variables were statistically significant at 5 per cent p-value: VIII region, XIII region, age, chronic disease, accident, work insurance, and the natural logarithm of out-of-pocket payment. Our results indicate that the odds of using health care are 2. Additionally, the odds of using health care services for individuals with chronic conditions are Accidents increase the odds of using health care by Finally, increases in the natural logarithm of out-of-pocket payment were associated with increasing the odds of using health care services by However, one needs to be cautious in interpreting this result because we used cross-sectional data, and the possibility of reverse causality, i.
For the second part of the model, estimated coefficients of the variables work status and natural logarithm of out-of-pocket payment were statistically significant. Therefore, having work is associated with a reduction of the use of health care services by Additionally, as the natural logarithm of out-of-pocket payment increases by 1 per cent the frequency of use of health care services is associated with an increase of 1.
Interpretation of this result, however, should be cautious for the potential reverse causality. The relationship between types of health insurance and healthcare use was not statistically significant. Distribution of actual , predicted and standardized need for health care services by income quintile. Distributions of actual need - predicted and need - standardized health care use for year The poorest 20 per cent used on average 7.
After we standardized the values the wealthiest people used almost twice the health care as the poor people did. Decomposition of concentration index for access to health care use , Non - need factors. The contribution of all need factors is negative, indicating that if utilization were determined by need alone, it would be pro-poor. The aggregate contribution of all need factors is about Logarithm of household expenditures and health insurance coverage increases the concentration index by approximately The residual difference between the unstandardized concentration index and the sum of the contributions of all need and non-need factors is larger for the partial effects probit approach, mainly because this gives a slightly larger estimate of the contribution of household expenditure.
Moreover, a positive horizontal inequity index indicates that the better-off make a greater use of health care services in Chile. The analysis of the distribution of actual need, predicted need and standardized need, and also the decomposition index all provide evidence of pro-rich inequities in the use of medical care.
Equity in health care utilization in Chile
Moreover, our results indicate that the poor are using less health care services than expected according to their needs. This analysis of horizontal inequities in health care utilization supplement our two-part model analysis that focusing on variables affecting utilization of health care utilization. Such equity analysis produces important information for policy concern of equity in health care utilization, independent of two-part model analysis.
Societies that concern equity of health care utilization will need to conduct similar equity analysis to provide evidence for policy assessment and recommendations. The two-part model estimation indicated some key factors that are affecting utilization of health care services. We find that the major predictors of service utilization between and using the CASEN dataset are education, the implementation of AUGE program and the type of health care payment.
This analysis provides important evidence of the achievements of the AUGE program, which, according to our results, increased Chilean's utilization of health care services. At the same time, while AUGE increased the overall utilization of health care services in Chile, it was still fell short of achieving equity.
Our estimation in this study indicated that after AUGE was implemented, the utilization of health care in Chilean health care system is still pro-rich. That suggests either further policies are needed to improve equity of health care utilization in Chile, or the need to revise or modify AUGE in order to improve equity in health care utilization.
Results from the analysis of the Satisfaction and Out-of-Pocket payments survey for the year suggest that the strongest predictors of health care use include work status and out-of-pocket payment. Results from these two surveys support that health care payment is an important variable to assess health care use.
However, interpretation of the relationship between health care utilization and out-of-pocket payment should be cautious. As mentioned in the last section, due to our use of cross-sectional data, reverse causality could potentially exist. The estimated coefficient of the variable AUGE is not statistically significant for the year ; however, a possible explanation is that AUGE had just started in In another study conducted in Chile analyzing the use of medical services, the authors concluded that AUGE reform was not necessarily improved equity in the use of health care services, and that there are still barriers to achieve the equitable use of health care services [ 43 ].
They also found pro-poor distribution in the use of emergency room visits and hospitalizations. There are several limitations of this study related to the use of secondary data. The first one is a longer self-reported recall period of one year or six months for most of the questions related to health care in the CASEN survey, which may increase recall bias. Also, we are aware that estimates of health care use can suffer from the same recall bias. We also recognize that self-report bias might exist for variables such as service utilization and income.
Individuals tend to under-report their income; which may lead to underestimation of inequalities across income groups. However service utilization can either be under or over reported, therefore, the present analysis may be biased but it is uncertain of the direction. Some researchers believe that self-reporting of physician visits may be unreliable [ 45 ].
Sometimes underreporting occurs in service utilization and is likely to increase as utilization of services increases [ 46 ]. If this is the case, then estimates of inequity can be underestimated in this study. Furthermore, in our model specification and estimation, there is potential bias as a result of omitted variables that were not included in the explanatory variables. There are two possible omitted variables in our model estimation. The first one is availability of providers in community, which could be measured in terms of travel distance. Holding every other variables constant, availability of providers could have an impact on the utilization of health care.