A theoretical framework of the good health status of Jamaicans: using econometric analysis to model good health status over the life course

Abstract

Background:

In recent times, the World Health Organization has increasing drawn attention to the pivotal role of social conditions in determining health status. The non-biological factors produced inequalities in health and need to be considered in health development. In spite of this, extensive review of health Caribbean revealed that no study has examined health status over the life course of Jamaicans. With the value of research in public health, this study is timely and will add value to understand the elderly, middle age and young adults in Jamaica.

Objective:

The aim of this study is to develop models that can be used to examine (or evaluate) health of Jamaicans, elderly, middle age and young adults.

Method:

The current study used data from a cross-sectional survey which was conducted between July and October 2002. Stratified random probability sampling technique was used to collect the data from 25,018 respondents across the island. The non-response rate for the survey was 29.7% with 20.5% who did not respond to particular questions, 9.0% did not participated in the survey and another 0.2% was rejected due to data cleaning. Logistic regression analyses were used to model health status of Jamaicans, young adults, middle age adults and elderly. The predictive power of the model was tested using Omnibus Test of Model and Hosmer and Lemeshow (24) was used to examine goodness of fit of the model. The correlation matrix was examined in order to ascertain whether autocorrelation (or multi-collinearity) existed between variables.

Results:

Using logistic regression analysis, eleven variables emerged as statistically significant predictors of current good health Status of Jamaicans (p<0.05). The factors are retirement income (95%CI=0.487-0.958), logged medical expenditure (95% Confidence Interval, CI =0.907-0.993), marital status (Separated or widowed or divorced: 95%CI=0.309-0.464; married: 95%CI=0.495-0.667; Never married), health insurance (95%CI=0.029-0.046), area of residence (other towns:, 95%CI=1.052-1.455; rural area:), education (secondary: 95%CI=1.167-1.576; tertiary: 95%CI=1.466-2.820; primary or below: OR=1.00), social support (95%CI=0.745-0.964), gender (95%CI=1.281-1.706), psychological affective conditions (negative affective: 95%CI=0.939-0.980; positive affective: 95%CI:1.047-1.107), number of males in household (95%CI:1.066-1.235), number of children in household (95%CI=1.117-1.266) and previous health status.

Conclusion:

The study concludes that good health status across the three age cohorts can be modelled using data for Jamaicans. Health status is determined by a number of non-biological factors, and that poor health status is difficult to model as a low proportion of the data was correctly classified. Public health requires research with which to make more informed decisions, which means that this study offers an understanding of Jamaicans as well as young adults; middle aged adults and elderly.

Introduction

Health is a multidimensional construct which goes beyond dysfunctions (illnesses, ailment or injuries)[1–14]. Although World Health Organization (WHO) began this broaden conceptual framework in the late 1940s[1], Engel[3] was the first to develop the biopsychosocial model that can be used to examine and treat health of mentally ill patient. Engel's biopsychosocial model was both in keeping with WHO's perspective of health and again a conceptual model of health. Both WHO and Engel's works were considered by some scholar as too broad and as such difficult to measure[15]; although this perspective has some merit, scholars have ventured into using different proxy to evaluate the ideal conceptual definition forwarded by WHO for some time now. Psychologists have argued that the use of diseases to proxy health is unidirectional (or negative)[2], and that the inclusion of social, economic and psychological conditions in health is broader and more in keeping with the WHO's definition of health than diseases. Diener was the first psychologist to forward the use of happiness to proxy health (or wellbeing) of an individual[16,17]. Instead of debating along the traditional cosmology health, Diener took the discussion into subjective wellbeing. He opined that happiness is a good proxy for subjective wellbeing of a person, and embedded therein is wider scope for health than diseases. Unlike classical economists who developed Gross Domestic Product per capita (GDP) to examine standard of living (or objective wellbeing) of people as well this being an indicator of health status along with other indicators such as life expectancy, Diener and others believe that people are the best judges of their state. This is no longer a debate, as some economists have used happiness as a proxy of health and wellbeing[18–20]; and they argued that it is a good measurement tool of the concept.

Whether the proxy of health (or wellbeing) is happiness, self-reported health status, self-rated health conditions, life satisfaction or ill-being, it was not until in the 1970s that econometric analyses were employed to the study of health. Grossman[9] used econometric to capture factors that simultaneously determine health stock of a population. Grossman's work transformed the conceptual framework outlined by WHO and Engel to a theoretical framework for the study of health. Using data for the world, Grossman established an econometric model that captures determinants of health. The model read (Model 1):

H_t= f (H_t-1, G_o, B_t, MC_t, ED) Model (1)

where Ht – current health in time period t, stock of health (Ht-1) in previous period, Bt – smoking and excessive drinking, and good personal health behaviours (including exercise – Go), MCt,– use of medical care, education of each family member (ED), and all sources of household income (including current income).

Grossman's model was good at the time; however, one of the drawbacks to this model was the fact that some crucible factors were omitted by the aforementioned model. Based on that limitation, using literature, Smith and Kington[10] refined, expanded and modified Grossman's work as it omitted important variables such as price of other inputs and family background or genetic endowment which are crucible to health status. They refined Grossman's work to include socioeconomic variables as well as some other factors [(Model 2)].

H_t= H* (H_t-1, P_mc, P_o, ED, E_t, R_t, A_t, G_o) Model(2)

Model (2) expresses current health status Ht as a function of stock of health (Ht-1), price of medical care Pmc, the price of other inputs Po, education of each family member (ED), all sources of household income (Et), family background or genetic endowments (Go), retirement related income (Rt ), asset income (At).

It is Grossman's work that accounts for economists like Veenhoven's[20] and Easterlin's[19] works that used econometric analysis to model factors that determine subjective wellbeing. Like Veenhoven[20], Easterlin[19] and Smith and Kington[10], Hambleton et al.[6] used the same theoretical framework developed by Grossman to examine determinants of health of elderly (ages 65+ years) in Barbados. Hambleton et al.'s work refined the work of Grossman and added some different factors such as geriatric depression index; past and current nutrition; crowding; number of children living outside of household; and living alone. Unlike Grossman's study, he found that current disease conditions accounted for 67.2% of the explained variation in health status of elderly Barbadians, with life style risks factors accounting for 14.2%, and social factors 18.6%. One of the additions to Grossman's work based on Hambleton et al.'s study was actual proportion of each factor on health status and life style risk factors.

A study published in 2004, using life satisfaction and psychological wellbeing to proxy wellbeing of 2,580 Jamaicans, Hutchinson et al.[21] employed the principles in econometric analysis to examine social and health factors of Jamaicans. Other studies conducted by Bourne on different groups and sub-groups of the Jamaican population have equally used the principles of econometric analysis to determine factors that explain health, quality of life or wellbeing[5,8,22,23]. Despite the contribution of Hutchinson et al's and Bourne's works to the understanding of wellbeing, there is a gap in the literature on a theoretical framework explains good health status of the life course of Jamaicans. The current study will model predictors of good health status of Jamaicans as well as good health status of young adults, middle age adults and elderly in order to provide a better understanding of the factors that influence each cohort.

Material & Methods

Participants and questionnaire

The current research used a nationally cross-sectional survey of 25,018 respondents from the 14 parishes in Jamaica. The survey used stratified random probability sampling technique to draw the 25,018 respondents. The non-response rate for the survey was 29.7% with 20.5% who did not respond to particular questions, 9.0% did not participated in the survey and another 0.2% was rejected due to data cleaning. The study used secondary cross-sectional data from the Jamaica Survey of Living Conditions (JSLC). The JSLC was commissioned by the Planning Institute of Jamaica (PIOJ) and the Statistical Institute of Jamaica (STATIN). These two organizations are responsible for planning, data collection and policy guideline for Jamaica.

The JSLC is a self-administered questionnaire where respondents are asked to recall detailed information on particular activities. The questionnaire covers demographic variables, health, immunization of children 0 to 59 months, education, daily expenses, non-food consumption expenditure, housing conditions, inventory of durable goods, and social assistance. Interviewers are trained to collect the data from household members. The survey is conducted between April and July annually.

Model

The multivariate model used in this study is a modification of those of Grossman and Smith & Kington which captures the multi-dimensional concept of health, and health status. The present study further refine the two aforementioned works and in the process adds some new factors such as psychological conditions, crowding, house tenure, number of people per household and a deconstruction of the numbers by particular characteristics i.e. males, females and children (ages ≤ 14 years). Another fundamental difference of the current research and those of Grossman, and Smith and Kington is that it is area specific as it is focused on Jamaican residents.

The proposed model that this research seeks to evaluate is displayed below [(Model 3)]:

H_t = f(H_t-1, P_mc, ED_i, R_t, A_t, Q_t, HH_t, C_i, En_i, MS_i, HI_i, HT_i, SS_i, LL_i,X_i, CR_i, D_i, O_i, ∑(NP_i,PP_i), M_i, N_i, FS_i, A_i, W_i, ε_i) Model (3)

The current health status of a Jamaica, Ht, is a function of 23 explanation variables, where Ht is current health status of person i, if good or above (i.e. no reported health conditions four week leading up to the survey period), 0 if poor (i.e. reported at least one health condition); Ht-1 is stock of health for previous period; lnPmc is logged cost of medical care of person i; EDi is educational level of person i, 1 if secondary, 1 if tertiary and the reference group is primary and below; Rt is retirement income of person i, 1 if receiving private and/or government pension, 0 if otherwise; HIi is health insurance coverage of person i, 1 if have a health insurance policy, 0 if otherwise; HTi is house tenure of person i, 1 if rent, 0 if squatted; Xi is gender of person i, 1 if female, 0 if male; CRi is crowding in the household of person i; ∑(NPi,PPi) NPi is the summation of all negative affective psychological conditions and PPi is the summation of all positive affective psychological conditions; Mi is number of male in household of person i and Fi is number of female in household of person i; Ai is the age of the person i and Ni is number of children in household of person i; LLi is living arrangement where 1= living with family members or relative, and 0=otherwise and social standing (or social class), Wi.

Statistical analysis

Statistical analyses were performed using Statistical Packages for the Social Sciences (SPSS) 16.0 software for Widows. A single hypothesis was tested, which was ‘health status of rural resident is a function of demographic, social, psychological and economic variables.’ The enter method in logistic regression was used to test the hypothesis in order to determine those factors that influence health status of rural residents if the dependent variable is a binary one; and linear multiple regression in the event the dependent variable was a normally distributed metric variable . The final model was established based on those variables that are statistically significant (ie. p< 0.05) – ie 95% confidence interval (CI), and all other variables were removed from the final model (p>0.05). Continuing, categorical variables were coded using the “dummy coding” scheme.

The predictive power of the model was tested using Omnibus Test of Model and Hosmer and Lemeshow[24] was used to examine goodness of fit of the model. The correlation matrix was examined in order to ascertain whether autocorrelation (or multi-collinearity) existed between variables. Cohen and Holliday[25] stated that correlation can be low/weak (0 to 0.39); moderate (0.4-0.69), or strong (0.7-1.0). This was used in this study to exclude (or allow) a variable in the model. Finally, Wald statistics was used to determine the magnitude (or contribution) of each statistically significant variables in comparison with the others, and the odds ratio (OR) for the interpreting each significant variables.

Results

Modelling Current Good Health Status of Jamaicans, Elderly, Middle Age and Young adults

Predictors of current Good Health Status of Jamaicans. Using logistic regression analyses, eleven variables emerged as statistically significant predictors of current good health status of Jamaicans (p<0.05, see Model 4). The factors are retirement income, logged medical expenditure, marital status, health insurance, area of residence, education, social support, gender, psychological affective conditions, number of males in household, number of children in household and previous health status (Table 1).

Table 1.

Good health status of Jamaicans by some explanatory variables

H_t = f (H_t-1, R_t, P_mc, ED_i, MS_i, HI_i, SS_i, ARi, Xi, ∑ (NP_i, PP_i), M_i, N_i, ε_i) Model (4)

The model [i.e., Model (4)] had statistically significant predictive power (χ² (27) =1860.639, p < 0.001; Hosmer and Lemeshow goodness of fit χ²=4.703, p = 0.789) and overall correctly classified 85.7% of the sample (correct classified 98.3% of cases of good health status and correctly classified 33.9% of cases of dysfunctions).

There was a moderately strong statistical correlation between age, marital status, education, retirement income, per capita income quintiles, property ownership, and so these were omitted from the initial model (i.e., model 3). Based on that fact, three age groups were classified (young adults, age 15 to 29 years; middle age adults, age 30 to 59 years; and elderly, age 60+ years) and the initial model was once again tested. There were some modifications of the initial model in keeping with the age group. For young adults the initial model was amended by excluding retirement income, property ownership, divorced, separated or widowed, number of children in household, and house tenure. The exclusion was based on the fact that more than 15% of cases missing in some categories and a high correlation between variables.

Predictors of current Good Health Status of elderly Jamaicans. From the logistic regression analyses that were used on the data, eight variables were found to be statistically significant in predicting good health Status of elderly Jamaicans (p < 0.5) (see Model 5). These factors were education, marital status, health insurance, area of residence, gender, psychological conditions, number of males in household, number of children in household and previous health status (see Table 2).

Table 2.

Good health status of elderly Jamaicans by some explanatory variables

H_t = f (H_t-1, ED_i, MS_i, HI_i, ARi, Xi, ∑ (PP_i), M_i, Ni, ε_i) Model (5)

The model had statistically significant predictive power (model χ² (27) =595.026, p < 0.001; Hosmer and Lemeshow goodness of fit χ²=5.736, p = 0.677) and overall correctly classified 75.5% of the sample (correctly classified 94.6% of cases of good or beyond health status and correct classified 44.7% of cases of dysfunctions).

Predictors of current Good Health Status of middle age Jamaicans. Using logistic regression, six variables emerged as statistical significant predictors of current good health status of middle age Jamaican (p < 0.05) (Model 6). These factors are logged medical expenditure, physical environment, health insurance, gender of respondents, psychological condition, number of children in household and previous health status (see Table 3).

Table 3.

Good health status of middle age Jamaicans by some explanatory variables

H_t = f (H_t-1, P_mc, En_i, HI_i, Xi, ∑ (NP_i), Ni, ε_i Model (6)

Based on table 3, the model had statistically significant predictive power (model χ² (27) =547.543, p < 0.001; Hosmer and Lemeshow goodness of fit χ²=4.318, p = 0.827) and overall correctly classified 87.2% of the sample (correctly classified 98.3% of cases of good or beyond health status and correct classified 28.2% of cases of dysfunctions).

Predictors of current Good Health Status of young adult in Jamaica. Using logistic regression, two variables emerged as statistically significant predictors of current good health status of young adults in Jamaica (p<0.05) (Model 7). These are health insurance coverage, psychological condition, social class and previous health status (Table 4).

Table 4.

Good health status of young adult Jamaicans by some explanatory variables

H_t = f (H_t-1, W_i, HI_i, ∑ (NP_i), ε_i) Model (7)

From table 3, the model had statistically significant predictive power (model χ² (19) =453.733, p < 0.001; Hosmer and Lemeshow goodness of fit χ²=5.185, p = 0.738) and overall correctly classified 92.6% of the sample (correctly classified 99.0% of cases of good or beyond health status and correct classified 28.2% of cases of dysfunctions).

Limitations to the Models

Good Health Status of Jamaicans [i.e., Model 4], elderly [i.e., Model 5], middle age adults [i.e., Model 6], and young adults [i.e., Model 7] are derivatives of Model (3). Good Health Status [i.e., Model 4 – Model 7] cannot be distinguished and tested over different time periods, person differential, and these are important components of good health.

H_t = f (H_t-1, R_t, P_mc, ED_i, MS_i, HI_i, SS_i, ARi, X_i, ∑ (NP_i, PP_i), M_i, N_i, ε_i) Model (4)

The current work is a major departure from Grossman's theoretical model as he assumed that factors affecting good health Status over the life course are the same, this study disagreed with this fundamental assumption. This study revealed that predictors of good health status are not necessarily the same across the life course, and differently from that of the general populace. Despite those critical findings, healthy time gained can increase good health status directly and indirectly but this cannot be examined by using a single cross-sectional study. Health does not remain constant over any specified period, and to assume that this is captured in age is to assume that good or bad health change over year (s). Health stock changes over short time intervals, and so must be incorporated within any health model.

People are different even across the same ethnicity, nationality, next of kin and socialization. This was not accounted for in the Grossman's or the current work, as this is one of the assumptions. Neither Grossman's study nor the current research recognized the importance of differences in individuals owing to culture, socialization and genetic composition. Each individual's is different even if that person's valuation for good health Status is the same as someone else who share similar characteristics. Hence, a variable P representing the individual should be introduced to this model in a parameter α (p). Secondly, the individual's good (or bad) health is different throughout the course of the year and so time is an important factor. Thus, the researcher is proposing the inclusion of a time dependent parameter in the model. Therefore, the general proposition for further studies is that the function should incorporate α (p, t) a parameter depending on the individual and time.

An unresolved assumption of this work which continues from Grossman's model is that people choose health stock so that desired health is equal to actual health. The current data cannot test this difference in the aforementioned health status and so the researcher recommends that future study to account for this disparity so we can identify factors of actual health and difference between the two models.

Discussions

This study has modelled current good status of Jamaicans. Defining health into two categories (i.e., good – not reported an acute or illness; or poor – reported illness or ailment), this study has found that using logistic regression health status can be modeled for Jamaicans. The findings revealed that the probability of predicting good health status of Jamaicans was 0.789, using eleven factors; and that approximately 86% of the data was correctly classified in this study. Continuing, in Model (4) approximately 98% of those who had reported good health status were correctly classified, suggesting that using logistic regression to examine good health status of the Jamaican population with the eleven factors that emerged is both a good predictive model and a good evaluate or current good health status of the Jamaican population. This is not the first study to examine current good health status in the Caribbean or even Jamaica [6,21–23], but that none of those works have established a general and sub-models of good health over the life course.

In Hambleton et al's work, the scholars identified the factors (i.e., historical, current, life style, diseases) and how much of health they explain (R2=38.2%). However, they did not examine the goodness of fit of the model or the correctness of fit of the data. Bourne's works [22,23] were similar to that of Hambleton et al's study, as his study identified more factors (psychological conditions; physical environment, number of children or males or females in household and social support) and had a greater explanatory power (adjusted r square = 0.459) but again the goodness of fit and correctness of fit of the data were omitted. Again this was the case in Hutchinson et al.'s research.

A major finding of this study is the fact that with the low classification of poor health, logistic regression cannot be used to examine poor health but good health. Given across the different age cohort examined in this study, logistic regression cannot be used to evaluate poor health as less than 30% of cases were correctly classified, suggesting that Jamaicans irrespective of their true poor health status compared to good health.

Although the current study has concluded that logistic regression can be used to model good health status and not poor health conditions across the age life cycle of Jamaicans, the factors identified that determine good health status for elderly, had the lowest goodness of fit (approximately 68%) while having the greatest explanatory power (R2= 35%). The findings also revealed low explanatory powers for young adults (R2=22.6%) and middle age adults (R2=23%), with latter having a greater goodness of fit for the data as this is owing to having more variables to determine good health. Another important finding is of the eleven factors that emerge to explain good health status of Jamaicans, when age cohorts were examine it was found that young adults had the least number of predictors (i.e., health insurance, social class and negative affective psychological conditions). This suggests that young adults’ social background and health insurance are important factors that determine their good health status and less of other determinants that affect the elderly and middle age adults. It should be noted that young adult is the only age cohort with which social standing is a determinant of good health. Even though the good health status model that emerged from this study is good, the low explanatory power indicates that young adults are unique and further study is needed on this group in order to better understand those factors that account for their good health.

Conclusions

The study concludes that good health status across the three age cohorts can be modelled using data for Jamaicans. Health status is determined by a number of non-biological factors, and that poor health status is difficult to model as a low proportion of the data was correctly classified. However, good health status can be corrected fitted using data for Jamaica, indicating the value of this study to understanding health status of the sampled population. In this study, apart from the identification of the variables for each age cohort and that of the general populace, some variables are the same across the three cohorts. Public health requires research with which to make more informed decisions, which means that this study offers an understanding of Jamaicans as well as young adults; middle aged adults and elderly.