Roads and Poverty

We now turn to the estimation of the effects of road development on poverty in rural Laos. Travelers in rural Laos cannot fail to be impressed by the low quality of the road system. It seems obvious that improving these roads could contribute to poverty reduction by improving poor peoples’ capacity to take advantage of the market economy. But by now much can poverty be reduced in this way?

4.1 The LECS data

The LECS surveys have been undertaken every 5 years since 1992-93:

- LECS 1 1992-93 - LECS 2 1997-98 - LECS 3 2002-03

The LECS 1 survey is different from the latter two, making comparison of its results with the later surveys hazardous. LECS 2 and 3 are quite similar and can be compared. The present study focuses on these two surveys.

The 1997-98 survey (LECS 2) covered 8,882 households containing 57,624 individuals. The data collection ran from March 1997 to February 1998 with about the same number of households (about 740) interviewed each month. The timing of the survey is important because as the discussion above indicates, LECS 2 was conducted at a time of high inflation, which reached annual rates well over 100 per cent. The data on consumption expenditures were collected in current prices, making the deflation of these expenditures into constant price terms particularly important. Of the 8,882 households covered, 6,874 were rural and the remaining 2,008 urban. In this study, only the data relating to rural households are used.

The 2002-03 survey (LECS 3) covered 8,092 households containing 49,790 individuals with the data collection extending from March 2002 to February 2003. Of these households 6,488 were rural and the remaining 1,604 were urban. In addition to data on expenditures, the LECS data include the following relevant variables (section codes of LECS in parentheses):

Province
District
Village

Characteristics of household
Number of adults
Number of members
Household consumption expenditure per person
Household income per person

Household ownership of assets
Irrigated land (B)
Dry land (B)
Rice husking machine (B)
Number of cows or buffaloes

Characteristics of household head
Age
Male (B)
Years of schooling
Unemployed (B)
Paid employee (B)
Employer (B)
Self-employed (B)
Farmer (B)
Unpaid family worker (B)
Outside labor force (B)

Educational characteristics of children in primary age group
Whether enrolled in school during past 12 months – C 5 (B)
If not, why not – C 6
Household expenditure on that child’s education – C 11
Distance from home to school attended – C 14
Time taken to travel to school – C 15

Health of household members
Whether treatment sought during last 4 weeks – D 7 (B)
Type of facility – D 9
Transport cost incurred in accessing the facility during the last 4 weeks – D 13

Village characteristics Electricity network (B)
Permanent market (B)
Scheduled passenger transport (B)
Distance to main road
Primary school (B)
Piped water or protected well (B)
Pharmacy (B)
Medical practitioner (B)
Trained nurse (B)
Community health worker (B)
Immunization program (B)
Urban (B)
Rural with access to road (B)
Rural without access to road (B)

It is important to note that these are sample surveys, not censuses. The number of households sampled is about 1.2 per cent of the total number of households within Lao PDR, and the individual households sampled in each survey are seldom the same. In any case, households are not identified individually and it is therefore not possible to compare the same households across LECS 2 and LECS 3.

It should be noted that “Distance to main road” is one of the variables listed, but this variable is known to be of unreliable quality, a point that is emphasized by data enumerators themselves. The variables “Rural with access to road” and “Rural without access to road” are considered more reliable and these are the data used in the present study. These variables reflect yes/no answers from households and are treated as dummy (0,1) variables in the regression analysis.

4.2 Analysis

It is convenient to move directly to the regressions that were estimated. Nominal consumption expenditures per household member were deflated to December 1999 prices using monthly provincial consumer price index data as summarized in Figure 9 [ PDF 135.5KB | 1 page ]. The deflation was conducted at a monthly level. This is especially important in the case of LECS 2, as noted above. The dependent variable was then the natural logarithm of real per capita expenditure.

The treatment of the dummy variables for dry season access to roads and wet season access needs explanation. We used dummy variables D and W, where D takes the value 0 if the household reports no dry season access and 1 if it reports road access. Then, W is defined similarly for wet season access. There was no household for which D was zero and W was 1. With respect to road access there were therefore three categories of households:

- (i) no road access at all: D = 0, W = 0, - (ii) access in dry season but not wet season: D = 1, W = 0, - (iii) access in both seasons: D = 1, W = 1.

The numbers of households belonging to each of these categories are summarized in Table 2 [ PDF 89.8KB | 1 page ]. In LECS 2, 31 per cent of households belonged to category (i) and this barely changed in LECS 3. These are the most isolated households of the country and according to these data little progress was made in providing them with road access over this period. In category (ii) – dry season access but not wet season access the proportion declined from 28 per cent in LECS 2 to 16 per cent in LECS 3. Thus the number of households which had wet season access as well as dry season access increased between these two surveys by 12 per cent of all households. In LECS 3, 52 per cent of all household had year-round road access.

The estimated regression equation handled this combination of outcomes through an interaction term. The right hand side variables thus included the terms

αD+ βD.W

where α and β are estimated coefficients. In case (i) above D and D.W are both 0. In case (ii) D = 1 and D.W = 0. In case (iii) D and D.W are both 1. The effect of dry season access alone is given by á and (noting that whenever W = 1, D = 1 also) the combined effect of dry and wet season access is given by α + β.

4.3 Regression results: LECS 2 and LECS 3

The regression results for LECS 2 and 3 are reported in Table 3 [ PDF 90KB | 1 page ] and Table 4 [ PDF 90.1KB | 1 page ]. In the case of the LECS 2 results the estimated coefficients had the expected signs, including the education variables and asset ownership variables, with the exception of “Not female head”, which had a negative but not significant sign. The variable “Reach dry” had the expected positive sign, but was not significant. The variable “Reach rain” had a positive and highly significant coefficient. According to these results, there was a high return to having wet season access in the LECS 2 data set.

The significance of this result for poverty incidence is explored in Figure 14 [ PDF 107.4KB | 1 page ] and Figure 15 [ PDF 149.2KB | 1 page ] and in Table 5 [ PDF 89.9KB | 1 page ]. Figure 14 [ PDF 107.4KB | 1 page ] shows the actual cumulative distribution of the logarithm of real consumption expenditures per person obtained from the LECS 2 data set. These data were assembled by calculating real consumption expenditures per person for all rural households, taking the natural logarithm and then sorting them from the lowest to the highest. The diagram also shows three estimated distributions, which use the regression results reported in Table 3 [ PDF 90KB | 1 page ], above.

P1. The predicted level of real expenditures using the actual values of the dummy variables D and W as observed in the data as well as actual values of all other independent variables. The difference between this prediction and the actual data is the error of the regression.

P2. The predicted level of real expenditure when all households have the value of D = 1 and W takes its values in the actual data, along with the actual values of all other independent variables.

P3. The predicted level of real expenditure when D = 1 and W = 1 for all households, along with the actual values of all other independent variables. The difference between P1 and P2 is an estimate of the degree to which real consumption expenditures could be increased if all households had access to roads in the dry season, but wet season access remained as observed in the data. The difference between this and P3 is then the degree to which real expenditures could be increased if all households had access to roads in the dry season and the wet season as well. Clearly, the difference between P1 and P3 indicates the potential for increasing real expenditures through road improvement.

The figure then uses these calculations to project levels of poverty incidence. In this exercise the poverty line is selected so that the predicted level of rural poverty incidence (P1 above) replicates the level of rural poverty incidence officially estimated for the LECS 2 data – 42.5 %. Because the estimated coefficient α is so small, the difference between the estimated level of poverty incidence in P1 and P2 is merely 0.06 per cent of the rural population (poverty incidence under P2 is 42.44%) and this small difference is not discernable in the diagram. But the difference between P3 and P2 is a further 7.58 per cent of the rural population (poverty incidence under P3 is 34.86%). This is the lower horizontal line in Figure 14 [ PDF 107.4KB | 1 page ] and Figure 15 [ PDF 149.2KB | 1 page ]. This number of rural people is equivalent to about 6 per cent of the total population of Lao PDR. According to these estimates, poverty incidence in Lao PDR could be reduced permanently by 6 per cent by providing all weather roads to all rural people.

It is notable that between the dates of LECS 2 and LECS 3, improved access to wet weather roads was indeed provided, as shown in Table 2, above. Fully 12 per cent of the rural population gained this form of access, compared with the 60 per cent of the same population that lacked it in 1997-98. This improvement was therefore about one fifth of the potential increase in wet season access. Interpolating linearly, the reduction in poverty incidence may therefore be estimated at about 1.2 per cent of the rural population. Rural poverty incidence actually declined by 9.5 per cent over this same period (Table 1 [ PDF 90.6KB | 1 page ]). Therefore these results imply that about 13 per cent (one sixth) of the reduction in rural poverty incidence that occurred between LECS 2 and LECS 3 can be attributed to improved wet season road access.

Turning to the LECS 3 results, Table 4 [ PDF 90.1KB | 1 page ] summarizes the regression results. The coefficient for dry season access is larger than for LECS 2 and more significant. The coefficient for wet season access, while still highly significant is now about two thirds of its value in LECS 2. The combined effect of providing dry and wet season access, the sum of these two coefficients, increased from 0.134 to 0.19.

These results may be interpreted as follows. The improvement in wet season access that occurred between LECS 2 and LECS 3 reduced somewhat the marginal return to providing wet season access, but it still remained large. Although there was no significant improvement in provision of dry season access between these two surveys, the increased market access available to households which had dry season access raised the real expenditure differential between those which did and those which did not have dry season access. This increase in market activity raised the real return to provision of road access.

Figure 16 [ PDF 169.3KB | 1 page ] and Figure 17 [ PDF 147.7KB | 1 page ] now show the implications of these results for predicted real expenditures, as previously, and Table 6 [ PDF 89.1KB | 1 page ] summarizes estimates of their implications for poverty incidence. Again, the poverty line is chosen such that the predicted level of poverty incidence replicates the preliminary World Bank estimate of rural poverty incidence based on LECS 3 of 33 % (See Table 1 [ PDF 90.6KB | 1 page ]). Official estimates have not yet been released. The three horizontal lines shown in each of Figure 16 [ PDF 169.3KB | 1 page ] and Figure 17 [ PDF 147.7KB | 1 page ] correspond to the levels of poverty incidence under P1 (33.00%, the top line), P2 (29.72%, the middle line) and P3 (25.90%, the lower line).

It should be noted that the World Bank estimates of rural poverty incidence for LECS 2 and LECS 3 (42.5% and 33%, respectively), when combined with the LECS 2 and LECS 3 survey data, imply poverty lines of 114,281 and 99,138 kip per person per month, respectively, when deflated by the consumer price index and expressed in December 1999 prices.¹ That is, the World Bank’s rural poverty lines increased in nominal terms somewhat less than the CPI. This outcome seems broadly consistent with the fact that the expenditures of the poor include larger shares of food than the non-poor, and (from Figure 8) the prices of food declined relative to those of non-food over this period.

According to our estimates, rural poverty could be reduced by 3.32 % (one tenth of the present number of the rural poor) if all rural households had dry season road access without any improvement in wet season access (the difference between P1 and P2). A further 3.77 per cent of the rural population could be raised from poverty if in addition all rural households had access to usable roads in the wet season as well. Together, if all rural households were provided with all-weather road access, poverty incidence in rural areas could be reduced by 7 per cent, equivalent to about 5.6 per cent of the total population of Lao PDR. This estimate is very close to that obtained from LECS 2.

4.4 Regression results: The change from LECS 2 and LECS 3

A possible objection to the analysis performed above is that it ignores the possible implications of the ‘endogenous placement’ problem. If improved roads were provided to better off areas, rather than independently of household real consumption, the relationship between better roads and real expenditures might not have the causal interpretation attributed to it in the above discussion.

This possibility was tested by assembling data on road improvement that occurred between LECS 2 and LECS 3. These data were assembled at the district level of which there are 140 in Lao PDR. These district level data are provided in Appendix A at the end of this paper. The data are not derived from LECS but from independent compilation of data from regional government offices and from the Ministry of Roads in Vientiane. Some judgment is involved in assessing whether roads were or were not ‘all weather’ and whether they were maintained. These judgments reflect the assessments of regional level officers of the Ministry of Roads.

The change in average real expenditures per capita between LECS 2 and LECS 3 was then related to the improvement or non-improvement of roads as captured in this data set. In the presentation of the results in Table 7 [ PDF 89.8KB | 1 page ], insignificant coefficients not related to road development have been dropped. The base level of real per capita expenditures in LECS 2 (1997-98) was significant and with a negative coefficient, meaning that better off households did less well in proportional terms (the dependent variable is the change in the log of real expenditures) than poorer households. The base level of road access in 1997-98 was less important in explaining the improvement in average real consumption expenditures at the district level than the change in road access, where the coefficient was highly significant and numerically of similar magnitude to the value obtained from the cross sectional results.

A further, more direct, test of the endogenous placement problem was conducted by regressing the change in road access that occurred between LECS 2 and 3 on the level of initial real per capita expenditure in LECS 2. The regression was done using regional level observations by taking the means of the district level dummy variables for improved road access for each district within the region and regressing this on the regional means of the district level real per capita expenditure as recorded in LECS 2. If better off areas received preferential treatment in road improvement a significant and positive coefficient would be expected. The estimated coefficient was negative but insignificant.

These results are supportive of the findings of the cross-sectional analysis reported above, confirming that improved road access raises real consumption expenditures and thereby reduces poverty.

Download this Discussion Paper [ PDF 314.5KB| 42 pages ].

Comment(s)

There are [3] comment(s) for this entry. Post a comment.

1. Ahmad Raza Khan
   (posted 28 November 2008 / 02:15:56 PM)
   
   It is very impressive and one can learn from the report and implement these findings in our own country. I am from Pakistan where the major portion of population lives in rural areas. The road network in Pakistan is not impressive especially in rural areas the situation is even more worse.
   by building road infrastructure and improving the alread existing , we can change the lives of people.
   
2. Arounyadeth
   (posted 22 October 2008 / 08:38:37 PM)
   
   Road is very important for development in Laos as many studies suggested. However, the project from ADB or other donors mostly concentrate on East-West road network. So far, North-South road network has been mostly untapped, except some maintenance project for route 13. However, the study on direct link road from North-South should be studied, because North-South regions hold comparative advantages, which they could trade to each other, e.g. livestock and paddy. I really hope that there will be more studying on this matter.
3. Karel Martens
   (posted 29 June 2007 / 05:44:52 PM)
   
   It is important to note that roads will only alleviate poverty if it goes together with a transport service that is accessible for the poor. The expectation that roads alleviate poverty is based on the assumption that some form of para-transit will be available to travel with or without goods.
   In current circumstances, such services will most likely be provided by the market. However, as affluence increases, so will car ownership, and the same roads build to alleviate poverty, may become instruments to increase the gaps between those with and those without cars.
   it is therefore important to stress what is alleviating poverty: not the roads as such but the transport service it provides to the poor. By focusing on the transport service rather than roads the proper link between poverty alleviation and accessibility is made.