a. Need variables

A list of the disabilities and chronic conditions included is provided in additional file 1. We include all three of these variables in our standardizations (described below).

b. Use variables

We define two broad classes of household-level use variables (which are then scaled to household size to get per-capita measures):

1. An indicator variable measuring the presence of any health related expense, including spending on health insurance.

2. A variable measuring the dollar amounts of health care spending.

The reason we are able to use the presence of any health related expense as an indicator of use is that no health care financing plans in Lebanon involves complete coverage of health expenses, with the exception of the government provided insurance plan for military and security personnel. For all other Lebanese citizens, even the most generous coverage involves some out-of-pocket expenditures, so we take the presence of health-related spending as an indicator of the use of healthcare services [9].

c. Catastrophic health payments

We then look at the incidence of catastrophic health payments (payments over 25 percent of expenditures per adult equivalent) across different population groups and socioeconomic characteristics.

d. Insurance coverage

We identify two broad classes of insurance plans: publicly provided plans (the National Social Security Fund, or NSSF, the Civil Servants' Cooperatives, municipal government plans, and plans of the security and armed forces) and privately provided plans (for the employed, the self-employed and the syndicated).

a. Means and concentration indexes

The methods that are employed to quantify the degree of equity in health care include descriptive and regression techniques using national household survey data described above. Descriptive methods include comparison of means and the 'concentration index' technique. The concentration index is a measure of how equally a health variable is distributed across a population ranked by income level. As a single numeric, the concentration index allows degrees of equity to be easily captured and compared, in order to determine the extent and nature of policy reform that is necessary.

b. Standardization: direct, indirect

A proper assessment of the equity of healthcare must control through regression analysis for the confounding effects of need and demographics, as well as other sources of heterogeneity that might affect healthcare use. We standardize our measures of the use of healthcare on three measures of need, as well as the main demographics we take to be correlated to utilization, using both the direct and indirect standardization techniques [8].

Direct standardization predicts the distribution of use by expenditure quintile that would be observed if the distribution of the confounding variables (need, age and sex) was the same for each quintile, but confounding variables had quintile-specific effects. Indirect standardization involves predicting the value of use in the same way, using the observed values for the confounding variables, but constraining their effect on use to be the same across all quintiles. In standardizing, we control for health confounding effects such as age, gender, and non-confounding effects such as education, the log of household expenditures and employment.

c. Methodology in poverty analysis

Our analysis of catastrophic health payments involves the calculation of headcounts and overshoots. The un-weighted headcount treats all households equally in calculating the share of households that incur a catastrophic health payment. This statistic assumes constant marginal utility of income. A measure that is more sensitive to equity concerns and that assigns more weight to households at the bottom of the distribution is the rank-weighted headcount. The concentration index measures the degree of equity in the incidence of catastrophic health payments across the income distribution and a negative index is indication that households at the lower end of the distribution of total household expenditures have a greater tendency to exceed the spending threshold on health and incur catastrophic health payments [7].

The depth of the impact of catastrophic payments is calculated using the overshoot, which calculates the average excess above the threshold. Like the headcount, the un-weighted version of the overshoot treats any dollar in excess of the threshold equally, regardless of which household is spending it. The rank-weighted overshoot instead assigns greater weight to overshoot spending by households at the bottom of the expenditure distribution. The concentration index measures the equity of the overshoot across the distribution of household expenditures: a positive value indicates that the overshoot tends to be greater among the better off.

When it comes to the analysis of poverty, we are unable to capture the impoverishing effects of healthcare spending (particularly when its scale is catastrophic) when our poverty line is defined on the basis of expenditure levels: the larger an individual's spending on healthcare, the likelier the individual's overall expenditures exceed the poverty line, which leads to an artificially lower poverty headcount. One approach that is commonly used to address this difficulty in constructing a valid measure of poverty is to calculate net poverty rates and net poverty gaps, which include only non-health spending [7]. The weakness of this approach is that insofar as the poverty line is constructed to include spending on healthcare, net figures will overstate poverty. Another approach that we propose is to extrapolate from the overall poverty line and the composition of the mean household's spending a poverty line for non-health expenditures. Poverty rates and gaps calculated based on this line are not affected by the extent of healthcare spending, nor should the threshold itself include expenditures on health. This method is similar in essence to the one developed by Wagstaff et al. (chapter 19) [7]. One of the major limitations in this approach is the arbitrariness of choosing the breakdown in the expenditures of the mean household to extract from the poverty line the amount of spending on non-health related goods.

a. The incidence of health-related expenses

We note the remarkable difference between the indirectly standardized measures and the observed directly standardized rates of healthcare use. Thus, when overall means of the non-confounding variables are used (in indirect standardization), predicted healthcare spending is much lower for all five quintiles than if we assumed quintile means for household expenditures, employment and education. The absolute difference in predicted usage rates across quintiles is roughly the same for all three measures, which makes the difference proportionately much larger when use is indirectly standardized. This is the result of the substantial differences in the non-confounding variables (expenditures, employment and education) across quintiles.

b. Out-of-pocket expenditures on health

To further examine the equity of average out-of-pocket expenditures on health, we construct concentration curves (Figure 2), which show the cumulative share of healthcare spending against the distribution of expenditures per adult equivalent. The curve shows a substantial gap from the equality line which is also reflected in the corresponding concentration index of 0.676. Out-of-pocket health expenditures appear to be progressive.

We reconstruct the concentration curve for out-of-pocket expenditures once they have been indirectly standardized on need (Figure 3). Indirectly standardized expenditures appear to be substantially more regressive than observed health expenditures: once need and relevant demographics are accounted for, the resulting measure of out-of-pocket expenditures on healthcare shows serious inequity in the healthcare spending across the income distribution.

c. The disability card

The use of healthcare services, whether measured by the presence of any health-related expense, the use of a disability card to access government provided health services, or the value of out-of-pocket expenditures on health appears to be regressive when measured against the distribution of overall household expenditures per adult equivalent. This result holds for the observed values of healthcare use as well as values standardized on need and other health determinants. Our results for equity in need showed disparity in the patterns of each need variable, but the more prevalent of our two need variables (the presence of a chronic health condition) put upper expenditure quintiles at more of a disadvantage. The results we find for use, even when standardized for need show patterns that consistently favor the rich.

a. Who incurs catastrophic health payments?

Unsurprisingly, the gender breakdown of individuals making catastrophic health payments (Figure 4) is strongly skewed towards women (where 53% of people incurring catastrophic health payments are women, whereas women are only 50% of individuals not making catastrophic health payments). The difference in the incidence of catastrophic payments by gender is significant at the 1% level.

We compare the geographical breakdown of households making catastrophic healthcare payments to that of households whose expenditure on healthcare does not exceed 25% of their overall household expenditures (Figure 5): the sharpest difference in the distribution is in the mohafaza of Nabatieh which accounts for around 9% of households whose healthcare payments fall below 25% of total household expenditures but close to 35% of households incurring catastrophic healthcare. The mohafaza of the South also shows sharp contrast: around 13% of households with non-catastrophic payments and close to 18% of households straddled with healthcare payments. Other remarkable differences that deserve note are that about 35% of households who spend less than a quarter of their income on healthcare are in Mount Lebanon. Similarly, the otherwise quite poor mohafaza of the North houses 17% of households whose payments are non-catastrophic but just 6% of households making catastrophic health payments. These results show an overlap between the districts that exhibit the starkest contrast in catastrophic healthcare payments and the incidence of repeated episodes of political violence (mainly in the governorates of Nabatieh and South Lebanon, in the southern part of the country) and their legacy of liabilities in terms of lifetime disabilities and risks in terms of unexploded mines and munitions.

We also look at the incidence of catastrophic payments per mohafaza (Figure 6) and find that Nabatieh has, by far, the highest risk of catastrophic health payments (17%), the South is a distant second (7%).

b. Welfare effects of catastrophic health payments

Table 10 shows that the concentration index of headcounts is consistently negative, indicating that households at the bottom of the distribution are more likely to incur catastrophic health payments, using a variety of different possible thresholds for catastrophic payments. This might be the result of a rigid payment structure that is not a function of income. The concentration index is greatest for the lowest value of the threshold, which thereby results in the biggest discrepancy between the weighted and un-weighted measures of the headcount.

Interestingly, however, the concentration index of overshoots is positive indicating that while poorer households are more at risk of catastrophic payments, the dollar amounts of these payments is progressive: the better off tend to face larger overshoots. This result may be driven by non-linearity in health consumption which requires a minimal level of health spending regardless of total expenditure, but then increases more than proportionately as total expenditures increase.

Using only non-food expenditures (and adjusting the thresholds accordingly) gives qualitatively similar results in terms of the analysis of progressivity and poverty impact: head counts have a negative concentration index which makes the discrepancy between un-weighted and weighted headcounts largest when the threshold is lowest. Overshoots are progressive.

a. Poverty analysis using national and World Bank poverty lines

Conventional poverty analysis is conducted using 4 different poverty lines: a deep national poverty line equivalent to expenditures of $2.2 per person per day, a national poverty line of $4 per person per day, the deep poverty line of $1 (or 1.08 in PPP) and the poverty line of $2 (or 2.15 in PPP) used by the World Bank, keeping in mind that the rate of exchange of the dollar to the Lebanese Pound is LL1,500/$. Poverty rates are calculated as a headcount of individuals lying below the poverty line, as a fraction of the overall population. Poverty gaps are calculated as the average shortfall from the poverty line (per year). And normalized poverty gaps express poverty gaps as a percentage of the poverty line. For poverty rates, poverty gaps and normalized poverty gaps, we calculate both gross and net measures, where health expenditures are included in the gross measures and netted out in the net measures.

b. Poverty analysis using a "non-health" poverty line

The spending profile on non-health related goods by the health-induced poor is comparable to that of individuals above the poverty line. It is the shortfall in their spending on health that pushes them into poverty.

In any analysis of the welfare effects of health payments, the health-obscured poor is a group that deserves attention if we are concerned about the validity of our methodology for measuring poverty: if we abstracted from healthcare payments, this group's spending profile on non-health goods would put them in poverty. So not only are the health-obscured trailing behind on non-health consumption, they are also spending more on healthcare than the typical "poor" household, which at once, obscures their poverty status and indicates that they are incurring large health expenses which is cause for concern in its own right. Thus, conventional poverty analysis fails to capture households whose poverty is health-obscured. To the extent that these households' spending on health hinders their ability to spend on other goods, health spending has an impoverishing effect on these households. While an-income based approach to poverty analysis would be able to detect the poverty of health impoverished households, this effect is missed by any expenditure based poverty analysis, as households whose health spending is substantial will show up as non-poor even when their spending on non-health goods is low.

a. Who is insured and what type of plan do they have?

We first examine the difference in the coverage rates across expenditure quintiles (Figure 7): only 18% of the poorest fifth are insured and the proportion rises monotonically across quintiles and reaches just over 70% for the richest 20% of the population.

Among the insured, Figure 7 shows that the breakdown across insurance schemes also changes across expenditure quintiles: publicly provided insurance is the dominant form of insurance for all but the top quintile and it represents an overwhelmingly large fraction of the insurance plans of the third quintile (close to 92%). Similarly, the rate of private coverage is around 5% of the insured in the bottom 60^th percentile, rises to 10% for the next quintile and accounts for a quarter of the insured in the top fifth of the expenditure distribution.

Figure 8 looks at the crossover of the poverty headcount with the insurance coverage variable, for both the gross and net poverty headcounts as well as the "non-health" poverty that we defined earlier: consistent with the results in Figure 7, Figure 8 shows that poverty significantly reduces insurance coverage for both private and public insurance. The difference between poor and non-poor is similar for both the gross and the net headcounts. The "non-health" poverty line defines sharper divides: the "non-health" poor have lower levels of all types of insurance than the poor under either of the other poverty lines. An interpretation of this result is that once we identify as poor people who appear above the conventional expenditure poverty line because of large expenditures on health, and we identify as non-poor people who fall under the conventional poverty line because they fall short of making "average" health expenditures, but spend more on non-health than we would expect them to at the poverty line, the residual pool of poor households seems to enjoy less insurance coverage.

The types of services that these plans cover differ substantially between privately and publicly provided insurance. Public plans offer the full range of services (medication, medical examination and lab analysis) whereas private plans only offer partial coverage of these services (Figure 9).

b. Hospitalization class and the quality of services

These results, however, should be seen in contrast with the quality of services provided by either class of plan. Public insurance accounts for the overwhelming majority of second and third class hospitalization service, whereas private insurance accounts for more first class hospitalization plans, by a relatively thin margin of 10% (Figure 10). The quality of care is obviously strongly related to hospitalization class however we have no real measure of the magnitude of the difference in the quality of service across classes.

The flip side of this analysis breaks down each insurance plan by class. We find that individuals with private insurance plans are concentrated in first and second class care (37% each) and only a quarter of them are eligible for third class hospitalization. Whereas only 5% of public insurance schemes provide first class hospitalization care, close to 60% provide second class care and the remaining 35% received third class care at hospitals (Figure 11).

Finally, we look at the differences in hospitalization class across quintiles: We break down the class coverage of each quintile to find that first class hospital care is only available to 2% of the lowest quintile and to 19% of the richest quintile. Second class care is also less available to the poorest (43%) than the richest (54%), but it is the next to richest quintile that benefits the most from second class care (60%). Access to third class care progressive, and the proportion of households with third class coverage decreases secularly as we move along the distribution of expenditures (Figure 12).

Figure 13 looks instead at the breakdown of each class of coverage over quintiles: the share of individuals coming from each of the first three expenditure quintiles increases monotonically as we move down hospitalization classes. Conversely, 67% of individuals with first class coverage are of the top expenditure quintile, while only 22% of individuals getting third class coverage are from the same quintile of expenditures.

Next we estimate a multinomial logit model of the class of hospitalization based on demographics (age, gender), insurance coverage (the presence of a plan, an indicator for publicly provided insurance), the presence of a chronic health condition, and the log of household expenditures. We reported the predicted proportion of each class of coverage based on these explanatory variables and find that the probability of coverage increases monotonically from the poorest to the richest quintile for both first and second class coverage, whereas the same probability decreases secularly from poorest to richest for third class coverage. For the bottom 60% of the expenditure distribution, the predicted proportion covered increases consistently from first to third class. Whereas the top 40% are likeliest to have access to second class hospitals care (Figure 14).

c. Health outcomes and health behavior by insurance plan

Next, we look at insurance coverage for people with chronic health conditions, across expenditure quintiles (Figure 15). We notice an interesting reversal: for the bottom two quintiles, individuals with a chronic health condition are likelier to be insured than individuals with no chronic diseases, whereas the reverse is true for people from the top 60% of the distribution of expenditures (although the difference is negligible for the top quintile). This is likely the effect of the composition of insurance types for each quintile: the richer quintiles are likelier to have private insurance, with premiums sensitive to the underlying health condition of the insured.

Expenditure quintiles

Beirut has the lowest concentration of the bottom two quintiles and by far the highest of the top quintile of expenditures. Mount Lebanon is also relatively well off with around half the population of the mohafaza in the top two fifths of the expenditure distribution. The poorest districts are by far the North and the South. 45% of the inhabitants of the North are in the bottom quintile of the distribution, a quarter in the next quintile and only 6.5% in the top quintile. Similarly, around a third of the residents of the South rank among the poorest fifth of the country and 60% are in the bottom two quintiles. The Bekaa is the province with the least inequality in the distribution of expenditures (Figure 16).

Need Variables

Use Variables

Insurance coverage shows large geographical disparities (Figure 17), with Beirut and Mount Lebanon showing 60% and 54% coverage respectively, but other mohafazas (namely the North, South and Nabatieh) trailing around 34%. The prevalence of private insurance is highest in Beirut and Mount Lebanon. Beirut has by far the highest incidence of private insurance (close to a third) and by far the lowest of public insurance (just over 60%). Mount Lebanon has the next highest rate of private insurance at around 17% and about three quarters of the insured in Mount Lebanon have publicly provided insurance. The North, the Bekaa and Nabatieh show similar patterns of rare private insurance (under 10%) and a large predominance of public insurance (close to or over 90%).

Hospitalization class also shows large disparities geographically (Figure 18): a fifth of the insured in Beirut have access to first class coverage and another two fifths have second class care. Only 4% of the insured in the South and 5% of those in the North have access to first class care. Second class care constitutes the bulk of coverage for all mohafazas, over 60% in the North and the Bekaa.