DYNAMICS OF HEALTH AMONG ADULTS IN SOUTH AFRICA

2 Background : This paper estimates trend of health mobility in South Africa using National Income 3 Dynamic Study (NIDS) and investigate whether the patterns of health mobility differs within 4 socioeconomic groups created by income and gender. Health is measured by SRHS, which 5 correlates with mortality and morbidity; thus, it is the best measure of health. 6 Methods: Using five waves of NIDS and various econometric models, this research estimates 7 health mobility in the period between 2007 and 2017. This study will use transition matrix as 8 descriptive analysis of health mobility and Conditional Maximum Likelihood Estimations to 9 analyse health mobility, trend of health mobility and relationship between health mobility and 10 health inequality within NIDS. 11 Results: The study shows that, among poor males, health mobility neither follows a health 12 selection or health constraint mobility trend; the high health mobility with ambiguous trends has 13 not decreased health inequality. Among the poor females, a negative health mobility trend is 14 observed; this research also found that health inequality has not creased. Among the non-poor 15 males, it is found that health mobility follows a gradient constraint trend which has decreased 16 health inequality. Among non-poor females, it is found that health mobility follows a health 17 selection trend which has not decreased health inequality. The results suggest that policy makers 18 should target both social determinants of health and health campaigns to deal with health 19 inequality among the poor males. 20 Conclusions: The trend of health mobility among poor females suggest that policy makers should 21 target the social determinants of health to combat health inequality. The trend of health mobility 22 among the non-poor males suggests that health mobility will eliminate health inequality. Lastly, 23 the trend of health mobility suggests that policymakers should target health campaigns to deal 24 with health inequality. 25


*Competing interests 19
There are no other relationships or activities that present a potential conflict of interest 20 *Funding 'Not applicable ' 21 *Authors' contributions 22 Usengimana Mutembereza had the idea of the paper and searched for appropriate data and the 23 research questions. He also discussed the model, run the models, too care interpretation, and 24 provided recommendations. 25 *Acknowledgements 'Not applicable ' 26 *Authors' information (optional) 27 Usengimana Mutembereza 28 Usengimana.shadrack@gmail.com 29 30 Introduction 1 Health mobility has implications for the health inequality (Umuhoza and Ataguba, 2018). The 2 people in excellent or very good health statuses can recover from the health shocks, while the 3 people in poor and fair health statuses have very little chances of recovering from health shocks; 4 because the latter's bodies have lower immunity to fight illnesses. Therefore, the South African 5 government has implemented a number of policies to lift the people out of bad health. The most 6 notable policy is the free primary healthcare at delivery, and some aspects of the tertiary 7 healthcare (Mayosi et al., 2012). However, health inequality continues persists as reported by 8 previous literature such as Obaku-Igwe (2015). This enigma has inspired this research to 9 investigate the patterns of health mobility in different Socio-Economic Statuses (SES). The results 10 of this research give indication on whether, in the long run, the current patterns of health mobility 11 will reduce the health inequality that exists in South Africa. 12 The previous research has studied the patterns of health mobility in the developed countries. For 13 example, Contoyannis et al. (2004) studies the health mobility in the UK; they found that on 14 average, among the British, the people that had previously reported poor health status had higher 15 probability of reporting different health status than the people that had reported excellent health 16 status. Their finding suggests that the patterns of health mobility will equalise health in the long 17 run for Britons. 18 However, such blessings are not for all countries. Previous literature reports that health mobility 19 has a parabolic relationship with the health inequality (Deaton, 2013); at the low levels of health 20 inequality, health inequality is declining as the health mobility increases; but at high levels of 21 health inequality, health mobility is associated with an increase in health inequality. The countries 22 with very high health inequality such as Russia, in early 2000, had an increasing health inequality 23 when health mobility increased (Heggebø, 2015 andBobak et al., 2000). health mobility and health inequality should be analysed in the lenses of the health gradient. They 26 both used the data from the Netherlands, which is an egalitarian country and they found that the 27 relationship between health inequality and health mobility is complex and there is a need to 28 investigate conditions upon which health mobility decreases health inequality. The previous 29 literature show that health mobility has different effect on health inequality in different groups of 30 people that are in the same country. Health mobility may also have different impact in a particular 1 group of people in different periods. 2 The negative relationship between health inequality and health mobility is either explained by 3 gradient constraint of health or health selection (Mutyambizi, et al., 2019 andMackbenbach 4 2012). The people in better health status have developed the network and lifestyle that keep them 5 healthy. They eat nutritious food and they abstain from unhealthy behaviour. While the people in 6 bad health have also developed unhealthy behaviour (Mutyambizi, et al., 2019). Health selection 7 hypothesis states that the people in bad health are not economically productive, and they are 8 deprived of the resources (Haas, 2006); the people in bad health will remain in bad health in the 9 long run. Therefore, when the relationship between health inequality and health mobility is driven 10 by health selection, policies that encourage change of behaviour need to be implemented for all 11 people to develop good behaviour. For example, sugar tax has been charged on beverages in 12 many countries including South Africa with a hope of decreasing sugar consumption. The 13 campaign and policies hope that people below the health threshold will move away from poor 14 health status or decreases the number of new cases of diabetes (Mutyambizi, et al., 2019). 15 In some countries, for example countries that have high inequality in access to healthcare, 16 increasing access to healthcare will increase health mobility for people previously reported poorer 17 health statuses more than for people that had previously reported better health statuses. In such 18 countries, mobility will decrease health inequality (Ro et al., 2016). This phenomenon is known 19 as gradient constraint (Elstad, 2001). The literature argues that when gradient constraint is 20 present, the determinants of health mobility should be boosted to eliminate health inequality in 21 the long run (Haas, 2006). 22 The literature on the gradient in South Africa has highlighted the issues that have kept a large 23 portion of the population in poor health and others in relatively better health statuses. Literature 24 has tested for the presence of the gradient constraint (Ro et al., 2016 to access social determinants of health. However, the policy interventions that do not include all 28 the people in ill-health will be futile in the long run. The people that currently have essential needs 29 will eventually be pulled into poverty (Marmot, 2017). 30 Therefore, there first objective of this research is to evaluate the patterns of health and the 1 implication of the patterns on health inequality. The investigation indicates if there is a need for 2 policy intervention and whether the intervention is to cover all people or first intervene for people 3 in ill-health that do not have essential determinants of health. In South Africa, there is limited 4 literature on the relationship between health mobility and health inequality. The literature has used 5 different approaches. For example, Lau and Ataguba (2015) used two waves of NIDS, they 6 estimate the relationship between social capital and health over the first two waves. The research 7 investigates the implied relationship between health inequality and health mobility and assumed 8 that health improves as the social capital increases. 9 Lau and Ataguba (2015) suggest that there has been a reduction in health inequality, because 10 the factors that are associated with health have improved. However, in South Africa, there is 11 evidence that social improvement does not imply a reduction in health inequality because  Therefore, the second objective for this research is to investigate the nature of observed 20 relationship between health inequality and health mobility. This study analyses the existence and 21 impact of health gradient and health threshold on health mobility to give insight into the 22 relationship between health mobility and health inequality. If health mobility has gradient 23 constraint, health inequality is decreasing because people in lower health status have higher 24 probability of improving health compared to people in higher health status. When health mobility 25 is driven by health selection hypothesis, health mobility has implication in the long run health 26 inequality; the people in higher health statuses have higher probability of improving their health 27 than the people in bad health statuses (Mutyambizi, et al., 2019;Deaton, 2013 andHaas, 2006). 28 The insights from this research are important. In 2012, the commission on development in South 29 Africa found that the nation will need to address the problem of health inequality for South Africa 30 to achieve the goals that set for national vision 2030. Furthermore, policymakers have applied 31 enormous amounts of effort to reduce the health inequality, but to a large degree, the results have 32 been disappointing. Among the developing countries, South Africa spends the largest portion of 1 her GDP (Obuaku-Igwe, 2015); however, the health indicators lag other developing countries. 2 This problem, large spending without desired results, is speculated to be linked to biases methods 3 used in research that guide the policy makers. These methods will be discussed under methods 4

section. 5
This research investigates the current health patterns of health mobility and assess if current 6 patterns can address the problem of health inequality in South Africa. The patterns of health 7 mobility in different SES groups indicate which group is having the greatest health mobility and 8 the analysis explores the gap in literature on health inequality. This is done by estimating the trend 9 of health mobility after controlling for social determinants of health. This allows the researcher to 10 indicate which SES group needs policy intervention. In addition, this insight enables the 11 researcher to recommend whether the policy makers should intervene within social determinants 12 of health or health behaviour campaigns. 13

Data 15
To achieve the objectives, this research uses the National Income Dynamic Study (NIDS) data 16 set. The data set currently has five waves available, which were collected starting from 2007 to  The data set has a number of questionnaires, there is a questionnaire for the adults, children, 23 proxy and the household; the adult questionnaire is answered by an individual that is above 15 24 years of age and lives in the household. The questionnaire for children and proxy questionnaire 25 are answered by an individual that is familiar with the particular individual. The household 26 questionnaire is answered by the oldest female in the house or any other person that is familiar 27 with the household spending. Therefore, this research uses the data collected through the adult 28 and household questionnaires. These are the only questionnaires where the individual answers 29 for themselves or for the household; this research is concerned with the self-reported health status 30 (Brown et al., 2012). Individual person has knowledge to assess their own health and the 1 assessment that they give is more accurate than the objective measures of health. 2 The table shows that the number of the people in the survey has been increasing over time. In to correct for people that have dropped out, prime-age males, for example. 13

14
The other cause for change in the numbers of people in the sample is attrition and retention. 15 Participating in NIDS is not enforced by law and some people may not be traced. Therefore, 16 information on some people, as the number of waves increase, is lost. If people drop out at a 17 certain pattern, the sample is no longer unbiased because the people that drop out of the survey 18 have a predictable health pattern. For example, if all the people dropped out of the survey were 19 in poor health status, the sample would have become biased; the population in poor health status 20 will be underrepresented. Therefore, there is a need to use the weight to adjust for the 21 underrepresentation (Ardington and Gasealahwe, 2012); this research investigates the nature of 22 attrition before analysing the relationship between health inequality and health mobility.  Therefore, the data set was split into poor males, poor females, non-poor males, and non-poor females. It 12 remains that the health variables in the previous wave does not determine who attrit in the current wave. 13 The results are not affected by individual heterogeneity, and likelihood-ratio tests show that panel-level 14 variance component is important. The results show that health attrition is random. Therefore, the NIDS data 15 set which is used in this research is appropriate to assess the individual health mobility; attrition does not 16 depend on the previous health status. Therefore, the analysis on dynamics of health using the SRHS 17 produces reliable results. 18 The literature has argued that SRHS is correlated with morbidity and mortality. Thus, it is an appropriate 19 measure of health because it is more reliable than objective measures of health. The literature observes that 20 a person that reports poor health status has higher probability of being dead in the next wave than a person 21 that had previously reported excellent health status (Shulman et al., 2006). SRHS  SRHS is the best measure of general health. When people report their health, they account for the past 10 condition of their health. 11 The SRHS was included in the Panel Study of Income Dynamics (PSID), which is among the prominent 12 dynamic data sets in the USA that started in the 1980s. However, the academics debated the reliability of 13 the variable for a decade. In mid 1990s, academics concluded that the SRHS is correlated with objective  Source: NIDS, table shows the sample size after all the restriction. 1 Therefore, inclusion of variable for initial health status would mitigate individual heterogeneity that is not 2 corrected by splitting the sample, because the scale that a person uses to report health status does not change 3 within the panel (Arellano and Bond, 1991). When a person experiences a decline/improvement in health, 4 they will report worse/better health than the previous reported health status. Therefore, initial health status 5 adjusts for unobserved heterogeneity, and splitting the sample adjust for systematic heterogeneity (Hauck 6 and Rice, 2004). The coefficients on the variable for previously reported health status and initially reported 7 health status are explained further in section: conditional maximum likelihood methods. 8 Descriptive analysis 9 The transition matrices are suitable to describe health mobility. The matrix analyses mobility between two 10 periods; it shows the probability of reporting health status reported in the initial wave (Chen and Cowell, 11 2017 Trede, 1999;Shorrocks, 1978 andPrais, 1955). The transition matrices 16 are also used to compare the health mobility between different SES groups (Shorrocks, 1978  ( 1 ) > ( 2 ), the society "1" has higher health mobility than the society "2". The mobility index shows 5 the likelihood of transiting from one health status to another status for SES group. The transition matrix 6 groups people together and assess their probability of moving between health statuses for the people in that 7 SES group. The mobility index is constraint between zero and one. Mobility index that is equal to one 8 indicates perfect health mobility and index that is equal to zero indicates perfect health immobility. The 9 comparison requires Shorrocks' mobility index as shown by Equation (1) (Shorrocks, 1978). Previously, 10 the index has been used to measure the income mobility and household poverty dynamics in South Africa 11 Klasen, 2005). 12 The poor males and non-poor males had mobility index of 0.913 and 0.884 respectively. The mobility was 17 calculated using the equation (1). This shows that the poor males had lower probability of reporting same 18 health status initially reported in wave five than non-poor males. In addition, the table shows the probability 19 of reporting same health status at wave five. For example, among the poor males that reported poor health 20 status in wave one, only 9.7% reported poor health status in wave five compared to 16.2 % for the non-poor 21 males. Both the poor males and non-poor males that initially reported poor health status have over 60% 22 probability of reporting either good, very good, or excellent health status. More than 50% of people that 23 initially reported good health status reported either very-good or excellent health status in wave five. The 24 people that initially reported better health status (very-good and excellent) had higher probability of 1 reporting same health status in wave five. 2 The transition matrices show that the health inequality between the poor and non-poor males have decreased 3 over the period of the study. The poor males have higher health mobility than the non-poor males, 4 particuraly the males in poor and fair health statuses. The transition matrices also show that, among both 5 the poor and non-poor males, the number of people that reported poor health has declined over the period 6 relative to other health statuses. This shows that the health inequality has decreased; the people that initially 7 reported poor health status have high probability of reporting a different health status in wave five. 8 Table 4 shows that probability of reporting a health status that is different from the health status reported 9 in wave one follows a positive gradient. The people that initialy reported poor health are less likely to report 10 same health status in wave five than people that initially reported excellent health status. The people initially 11 reporting excellent health status have lowest probability of reporting either poor or fair health status in wave 12 five. The table shows that health mobility is high, and health mobility has decreased health inequality over 13 the period of the study within both poor and non-poor males. 14  Poor females has slightly higher health mobility than non-poor females; poor female and non-poor female 19 had mobility index of 0.908 and 0.896, respectivelly. This indicate that the poor females and non-poor 20 females do not have major difference in their probability of changing their health status. This imply that 21 health inequality that existed, in wave one, between the poor and no-poor female has not decreased. The 22 transition matrices further show that health mobility increases as the people move from poor health status 23 to excellent health status. The people that initially reported poor or fair health statuses have high probability 24 of reporting a better health status. The probability of remaining in the initial health status increases as the 25 initially reported health status moves from poor health status to excellent health status. The female in 1 excellent health status have the highest probability of remaining in their initial health status. 2 Ataguba (2013) found that the poor people incur more illness than the non-poor. In addition, the poor people 3 have lower medical insurance coverage compared to the non-poor, therefore they have lower demand for 4 healthcare. These results that suggest that health mobility follows a positive gradient which has decreased 5 health inequality are not consistent with the previous literature. The inconsistency may be a result of biased 6 estimation of health mobility from transition matrix. obeys the initial conditions. The initial values must be fixed, which requires the initial values to be the 23 beginning of the series. The dependent variable must also have a common mean in different waves and the 24 initial values must not affect the latter health status. In addition, the initial values must be normally 25 distributed. The unobserved individual effect must be independent of the unobserved dynamic process so 26 that the process can converge towards a common mean. Lastly, the unobserved individual effect must be 27 independent of the unobserved dynamic process because initial value are random (Anderson and Hsiao, 28 1982). 29 These are strong assumptions, and SRHS fails to meet the requirements (Anderson and Hsiao, 1982). The 30 beginning of health cycle is unknown because the things that happen before a child is born have a bearing 31 on the adult's health. The SRHS does not converge to a common mean, the health is random, and it is 1 influenced by many factors (Deaton and Paxson, 1998). The initial health influences the latter health 2 statuses which violates initial condition. 3 Arellano and Bond (1991) confirm that the simple dynamic model cannot be used to study dynamics of 4 health. SRHS is a variable in micro-panel, which are naturally short. In the short panel, N →∞ and T → 5 Fixed number. Arellano and Bond (1991) found that a simple dynamic model on a short panel will produce 6 inconsistent results. 7 Wooldridge (2005) suggests that Conditional Maximum Likelihood Estimation (CMLE) deals with the 8 error term ( ) of the equation (2). The inclusion of the initial values of the dependent variable as the 9 explanatory variables transform the error term into an Independent and Identically Distributed (IID); this 10 process produces an error term that is comparable to the error term when the variables obeys the initial 11 conditions (Wooldridge, 2005 andHsiao, 1982). The error term contains the unobserved 12 individual heterogeneity and the process that controls for the individual heterogeneity makes the error term, 13 (ε i1 + v it ), normally distributed and eliminates unobserved heterogeneity (Wooldridge, 2005). 14 = SRHS it−1 + + + (2) 15 is the control variable such as age and financial standing at age of 15. 23 represent the error term, this is the variation in things that are not included in the model µ it in equation (3)  24 is the systematic error term, individual variation that is not controlled. ε i1 and v it are the random part of the 25 error term, while α i is the unobserved part of the systematic error term and it has a constant value. 26 Therefore, when both initial and lagged variable are added in the model, the unobserved part of the 27 systematic error term falls out because both the current and previous health status contain the same value 28 with different signs. CMLE controls for the initial condition (Wooldridge, 2005). The model produces 1 consistent results even though both the initial conditions and the asymptotes requirements maybe violated. 2 In the NIDS data set, SRHS takes the value of one if the person reports excellent health and five if the 3 person reports poor health. This study reverses the code from one for excellent to five and from five for 4 poor to one. This process simplifies the interpretation of the results. A positive coefficient indicates that the 5 people that previously reported excellent, very-good, good and fair health statuses have higher probability 6 of being in the same health statuses in current wave than the people that previously reported poor health 7 status which is the base category. While a negative sign indicates that the people that previously reported 8 excellent, very-good, good and fair health statuses have lower probability of reporting same health statuses 9 than the people that previously reported poor health status. 10 Gamma, γ, in equation (5) measure the relationship between current health and the previous health status. 11 The value of gamma is constrained between [-1, 1]. The significance and the size of the estimated 12 coefficients on the lagged categories of the dependent variable assess health mobility. Large and highly 13 statistically significant suggest that the current health has a strong relationship with previous health status. 14 Small and highly statistically significant suggests that the current health status has a weak relationship with 15 previous health status. Insignificant Coefficients show that the previous health has no relationship with 16 previous health status and significant coefficient shows that the current health is significantly related to the 17 past health (Contoyannis, et al., 2004). 18 The coefficient on the initial variable of health, φ, gives insight into the nature of health mobility. If 19 coefficient on initial health is statistically significant and lower than the coefficient on the corresponding 20 lagged health variable, health mobility has decreased health inequality (Contoyannis, et al., 2004). On the 21 contrary, if the coefficient on initial health is statistically significant and higher than the coefficient on the 22 corresponding lagged health variable, health mobility has not decreased health inequality. If coefficient on 23 initial health is not statistically significant and coefficient on the corresponding lagged health variable is 24 positive and significant, health mobility will decrease health inequality (Hauck and Rice, 2004). 25 This research control for various social factors that are associated with health because the analysis of the 26 dynamics of health that does not control for the social determinants of health is deemed unspecified which 27 causes the coefficients to be biased (Marmot, 2017). The social determinants of health as identified by the 28 literature are access to healthcare, livelihood environment, nutrition intake, social capital and individual 29 lifestyle. Table 6 shows the social determinants of health and how they are measured. 30 It is on the scale of 1 to 5 which assess the level violence exposure for the household.

Unhealthy behaviour
The person engages in gambling, smoking and alcohol consumption. The people with the highest ranking are involved in gambling, smoking and consume alcohol.

Control variables Education
It is continuous variable reporting highest education.

Log of mean income
The log of average of income per capita in five waves.

Log of real income
The log of real income per capita Employed 1 if the person is employed zero otherwise Unemployed 1 if the person is in labour market but not employed and zero otherwise Financial standing at 15 years Household financial standing at age of 15 years old. This is a constant value which is on a scale of 1 to 6. 1 is for the poorest families and 6 is the richest families. Urban Race Africans is base category other categories are Coloured, Indian and White Age The age in the first wave, age square and up to fourth power is included. Age doesn't change within the panel because change would be for everyone this can cause a bias in analysis.

Source: author. Literature that was consulted includes Omotoso and Koch, 2018 and Hauck and 1
Rice, 2004. 2

1
The results for the different SES groups are reported separately in Table 7. The models for men 2 and women are presented separately and further, in each gender group, the results for the poor 3 and non-poor are reported separately throughout. The separation control for systematic 4 heterogeneity which is reported between males and females, and poor and non-poor 5 (Wooldridge, 2005). 6 Table 7: Health mobility in each SES group   Table 7 shows that among poor males, the coefficient on the variables for previous health status is not 5 significant. Wald test show that the results are reliable for intepretation, because variables that are 6 theoretically correlated with health in South Africa are also included in the model. The people that have 7 previouslly reported excellent, very good, good and fair health statuses do not significantly have higher 8 probability of reporting same health statuses than people that have previously reported poor health status. 9 This suggest a high health mobility in the group of poor males. 10 It is worth noting that coefficients on variables for intitial excellent, very good and good health statuses are 11 statistically significant; each of the coefficients also have a high values which are more than 0.35. Only the 12 coefficient on variable for initial fair health status is not statistically significant ; the probability of reporting 13 fair health status for those initially reported fair health status is not significantly higher than for those 14 initially reporting poor health status. This suggest that, apart from people that initially reporting fair health 15 status, high health mobility that is observed will not decrease health inequality within poor males group in 1 a long-run. 2 Among poor females, all coefficiences on the previous health variables are statisticaally significant, and the 3 coefficients follow a limited negative health gradient. Wald test show that the results are reliable for 4 intepretation and the inclusion of the social determinants of health eliminate possibility of biased estimation. 5 Coeffient on variable for people that previously reported excellent health status is 0.316 and it is significant 6 at 5% significance level. The value of coefficient on the variable for people that previously reported very 7 good is lower than for excellent variable at 0.275 and significant at 5% significance level. However, the 8 coefficient on variable for people that previously reported good is higher than coefficient on variablefor 9 very good health at 0.295 and significant at 5% significance level. Coefficent on variable for people that 10 previously reported fair health status is the highest at 0.344 and significant at 5% significance level. The 11 result suggests that health mobility follows a negative health mobility apart from variable for those 12 previously reporting excellent health status. 13 It is noted that coefficients for the initial variables for health are smaller in magitude than coeffiences on 14 the previous health variables apart from the coefficient on the variable for people that previously reported 15 very good health status. This suggest that health mobility will decrease health inequality among the poor 16 females. However, health mobility will decrease health inequality at a rate that favours the poor females 17 that previously reported better health, which suggests that an aid that would move people from fair and poor 18 health status would accelerate health mobility and the rate at which health inequality decreases. 19 Among non-poor (both males and females), health gradient that favours the people that previously reported 20 fair health (positive gradient) is observed. The coefficients on the variables for previous health are 21 statistically significant and Wald test show that the results are reliable for intepretation. Since social 22 determinants of health are included, the results are not biased. The results show that the people that 23 previously reported better health are more likelly to report same health compared to people that previously 24 reported worse health status. 25 In group of non-poor males, the coefficient on variable for people previously reported excellent, very good, 26 good and fair health statuses are 0.437, 0.395, 0.308 and 0.229, respectively. These coefficients are 27 significantly higher than the value of corresponding coeffcients that are on the variables for people initially 28 reported the health status. This suggest that postive health mobility observed in non-poor males will 29 decrease health inequality in long-run. 30 Among the non-poor females, the coefficient on variables for people previous reported excellent, very good, 1 good and fair health statuses are 0.23, 0.22, 0.21, and 0.124, respectively. These coefficients are 2 significantly lower than corresponding coefficients on the variables for initially reported health status, 3 beside for fair health status. The results show that health mobility in non-poor females has not decreased 4 health inequality. 5 This research uses the quadrature to test the robustness of the results. When the number of quadrature 6 needed for the model to produce the result are changed, the results for all groups remain stable. Therefore, 7 current results are reliable because it is also noted that Rho statistics are lower than 0.1 for all the groups; 8 unobserved individual heterogeneity has low or no impact on the results. In addition, Wald test, Quadrature 9 test, Likelihood ratio test and attrition test show that these results are reliable. 10 Discussion of the results and conclusion. 11 The transition matrix shows that health mobility in all the groups follows a positive gradient constraint 12 pattern; people are likely to report a better health statuses in succeeding wave. The results suggest that 13 health mobility will decrease health inequality. Howoever, transitional methodology encounters a number 14 challenges that couldn't be addressed; the results might not be realiable, The second explaination of the health mobility that is observed among the poor men is their association or 3 health selection. The poor men might be using curative healthcare and not investing in preventative 4 healthcare which can explain high health mobility. If this is the explaination, then policy makers should 5 emphasise in health compaigns to alter the destructive behaviour. 6 Ataguba et al. (2011) found that South Africa represents a classic example of the inverse care law. 7 The healthcare usage decreases as the need for healthcare increases. Poor people and people that 8 report poor health status have tendencies to use the curative healthcare while non-poor and people 9 that report excellent health status have tendencies to use the preventative healthcare (Ataguba et al., 10 2011). This research find that the policy makers will need to intervene for the health inequality to decline. 11 However, this research find that it will require an innovative strategy. 12 The results for poor females have shown that health mobility follows a limited negative health gradient. 13 Limited negative health gradient suggests that poor females might experience health threshold on their 14 health mobility. The negative gradient is associated with a threshold in health mobility, if people in lower 15 health levels are unable to recover their health in long time (Mutyambizi, et al., 2019). At the first glance, 16 the results suggest that probability of being stuck in bad health status increases as the previous reported 17 health status decrease towards poor health status. 18 However, further investigation shows that, among poor females, health mobility has decreased health 19 inequality limited to people initially reported very good health status. The coefficient on the variable for 20 people initially reported fair or good health status are not statistically significant which indicate a high 21 health mobility over the period which dismiss the possibility of health threshold. The results show that 22 health selection can reduce health inequality in certain conditions. In South Africa, the use of Practical 23 Approach to Care Kit (PACK) has influenced poor women to use preventative healthcare which can explain 24 the decreasing health inequality that is observed in negative health gradient (Murdoch, et al., 2020). The 25 results also suggest that policy makers can increase health mobility if they increase access to social 26 determinants of health; reversing negative trend of health mobility would increase health mobility and 27 increase rate at which health inequality decreases. 28 The results for non-poor males show that health mobility follows gradient constraint. Health mobility in 29 this group is an ideal mobility beacause people that have previoyusly reported better health such as excellen 30 and very good have high probalility of reporting same health status while people that previously reported 1 bad health have high probability of reporting a better health status in current wave. Gradient constraint is a 2 result of access to social determinants of health which provide a protection against health shocks, and when 3 non-poor males get ill, they recover their health as shown by high health mobility among the people 4 previously reported lower health statuses because non-poor males have acess to both preventatitive and 5 curative healthcare (Harris, et al., 2011). 6 The results suggests that the health mobility has decreased health inequality over the period of the study. 7 The coefficients on the variable for previous health are greater in value than the coefficients on the variables 8 for initial health statuses. Non-poor males have established networks that enhance improvement in their 9 health and they have access to social determinants of health which enhances health gradient constraint. 10 The results for non-poor females show that health mobility follows a postive health gradient. Health 11 mobility increases as previous reported health status decreases from excellent health status to fair health 12 status. The results show a high health mobility between waves of the panel. The probability of reporting 13 same health status as the previous wave is lower for non-poor females than non-poor males. High health 14 mobility among non-poor females contradicts the expected results because womens are better examiners 15 of their health than men and are expected to use preventative healthcare (Harris, et al., 2011). Therefore, 16 compaigns would decrease the effects of health mobility that follows a health selection pattern. 17 The results show that health inequality, among non-poor females, has not decreased over the period; the 18 coefficients on the variables for previous health statuses are smaler in size than the coefficient on the 19 variables for initial health beside fair health status. The results show that health has high mobility, but health 20 mobility has not changed distribution of health in a long-run. Therefore, compaigns that encourages people 21 to join network and lifestyle that keep them health would enable health mobility to decrease health 22 inequality in a long-run (Pulsford et al., 2015). 23 This research finds that, among poor males, health mobility has no particular pattern and health 24 mobility does not decrease health inequality in long-term. Therefore, response from the policy 25 makers need to address the issue of health inequality through both social determinants and 26 campaigns. Among poor females, it is observed that health mobility follows a negative gradient 27 which does not decrease health inequality; the response needs to address health inequality through 28 social determinants. Among non-poor males, it is observed that health mobility follows a positive 29 gradient and health mobility decreases health inequality. Lastly, among non-poor females, health 30 mobility follows a positive gradient, but health mobility will not decrease health inequality; it is 1 suggested that health campaigns are needed to decrease the impact of health selection mobility.