Micro-level data on relative SEP of households and its dynamics in the context of an armed conflict have been presented. Our data stem from two cross-sectional household surveys, the first carried out just before, and the second one and a half years after the outbreak of an armed conflict in Côte d'Ivoire.
Our secondary data analysis revealed consistency of the underlying data, illustrated by the high recapture rate of households (89.7%), and the accuracy of population data (potential misreporting of only 0.5% of the original population in the second survey). The reliability of the data, methods, and the constructed wealth index was confirmed by the algebraic sign assigned to each asset. The only exceptions were the negative values assigned to the variables for house possession and land tenure. However, the absences of house and land ownership could be interpreted as proxies for employed work, which plausibly could be associated with higher wealth in the rural setting under investigation. In fact, households which did not possess land or housing were significantly more often engaged in non-agricultural work than other households (p < 0.001 for land and housing in both surveys). This exception has also been identified, and has been explained accordingly, in other rural settings when utilizing the same asset-based approach for wealth assessment (see for example reference ).
Despite the exceptional circumstances and the fact that 130 households (71.4%) mentioned socioeconomic difficulties, our analyses revealed weak socioeconomic dynamics with only every seventh household (14.3%) being labeled 'worse-off' or 'better-off' between the two surveys. Furthermore, no dramatic equity shifts were found. Total asset possession of all households varied little over the length of the study, with extremes of -15% for disposing of working cemented toilets and +13% for not having a toilet at all (see Tables 2 and 3 and endnote 4).
An attrition analysis of the 21 households which were interviewed in the first survey but could not be revisited in the second survey (see Figure 1) gave no indication of a selection bias, as the excluded households were symmetrically distributed over the wealthier and poorer wealth quintiles. Furthermore, there was no significant difference in the reported characteristics between those households included and those lost to follow-up. While we cannot exclude the possibility that households lost to follow-up lost all their assets or were banished as a consequence of the armed conflict, by the same token it is possible that they became wealthier or could improve their social status and decided to migrate for these reasons. Reports from those households that remained in the area and participated in the follow-up survey revealed that among households lost to follow-up, 5 out of 21 (23.8%) migrated to rebel-controlled areas and another 5 (23.8%) moved to government-controlled territories. This anecdotal evidence suggests that motivations and causes to migrate were diverse. A possible bias due to the operation of humanitarian aid projects in the area is also considered unlikely as the second round of household interviews took place before these humanitarian activities were fully implemented. Furthermore, the humanitarian aid projects were targeted on community rather than household or individual level. Hence, it is plausible that the observed dynamics are neither a matter of systematic asset in- or outflow nor a simple methodological artifact. Rather, the fluctuations in relative SEP seem to be the result of a limited reallocation of the always present assets among the studied households, without remarkable equity shifts.
The weak socioeconomic dynamics may at least partly explain why, except for the self-reported problems encountered since the beginning of the armed conflict, no significant associations were identified between the households' socioeconomic fate and other characteristics. As shown in Table 6, 'even' households mentioned significantly more often health-related problems than 'worse-off' or 'better-off' households. This finding is difficult to interpret, but the fact that 'better-off' households complained more often about the interruption of public services and the lack of food than their 'worse-off' or 'even' counterparts may reflect that they are not used to have any difficulties in these domains. Previous research in the same study area also observed that schoolchildren from wealthier households complained more often about suffering from disease symptoms than their less wealthy but equally healthy peers . This pattern was identified in other epidemiological settings as well [56, 57] and explained with higher expectations of the 'better-off', which makes them more sensitive to distress and consequently also more likely to complain.
Our findings of changed livelihood strategies are consistent with results from other reports [37, 44]. One study carried out in central, north and west Côte d'Ivoire investigated the effect of the same armed conflict on human resources and the functioning of the health system. Significant reductions of well-trained staff in both the public and the private sector were observed, along with a collapse of the health system and other components of public infrastructure . In the current investigation, we recognized a significant return to primary production of the interviewed heads of households. Nevertheless, we found no statistically significant associations between a household's socioeconomic fate and its main occupation or other important economic activities in the village of residence.
However, several particularities of the study design and the methodological approach applied may have influenced the outcome of the analyses. First, our sample size is small and therefore confidence intervals are large. This fact may partially explain why most changes and associations were insignificant. Second, due to the potentially low level of wealth before the outbreak of the armed conflict, even a small change in asset possession may have been sufficient for a change in the wealth quintile classification. Furthermore, a certain degree of elusiveness was inevitable as we had to rely on a secondary analysis of self-reported information and could not investigate all relevant indicators. For example, the insignificant associations between the households' socioeconomic fates and certain household characteristics may simply highlight that the usual characterization of households (e.g. age, sex, education, occupation, changes in household composition, accessibility or illness) does not capture the most important factors of what made households 'better-off' and 'worse-off' after the armed conflict. Local power structure, ethnic group, family feuds, and rivalry may be even more important than usual during an armed conflict, as implied by informal comments of the study participants regarding trafficking of drugs and other illegal and stolen goods (see endnote 6).
Certain weaknesses in the study design are partly owed to the very nature of the study, since working in conflict areas is a challenging task with many unforeseeable events. Access to, and movement of, populations may lead to a selection bias, as discussed above. Furthermore, war-related research cannot rely on any experimental design and instead must depend upon observational studies, which are limited in their ability to address causality. The timing of any war-related research is particularly difficult. Social tensions and sporadic physical violence may precede and only slowly wane after a fully fledged armed conflict. Hence, it is difficult to obtain reliable 'pre-conflict' baseline and 'post-conflict' follow-up data. Often, it is uncertain whether the observed dynamics were just the near completion of a process that already began years ago (i.e. before baseline survey) or whether many serious consequences might still occur in the future (i.e. after follow-up survey) (see endnote 1).
Methodologically, household asset-based approaches for estimating relative SEP have proven to be valid in rural Côte d'Ivoire, as well as in other African settings . This is especially true for the method applied here, which has been adapted and widely used for health surveys in various locations including the present study area [41, 42, 46–49]. Hence, the methodological approach seems to be adequate for analyzing the effects of an armed conflict on relative SEP of households. However, a study from southeast Nigeria found some indication that the reliability of asset-based wealth indices is only moderate , so we cannot exclude the possibility that a few of the identified changes in relative SEP were simply due to measurement errors.
Asset-based socioeconomic indices were calculated to estimate the households' relative SEPs in "a pragmatic response to data constraints" . Asset possession was measured only as a binary socioeconomic indicator, irrespective of quantity or quality. Assets were not valued at current monetary market prices, but weighted with 'raw' asset factors originating from PCA. These 'raw' asset factors were re-calculated for both surveys, which may seem quite arbitrary, even though they changed only slightly in our study. However, at times of an armed conflict, relative importance of assets may change rapidly. Therefore, we adopted an approach with flexible weights (see endnote 7). Nevertheless, computed household scores represented ordinal values, which allowed for ranking the involved households, as opposed to cardinal values (e.g. monetary values), which would also provide information about changes in absolute wealth.
The idea that "(asset) weights should be allowed to vary over time"  is in line with a statement by Sahn and Stifel (2000). However, in their aggregate analysis of poverty over time and across African countries, they used fixed weights based on the results of pooled wealth indicators. They evaluated socioeconomic dynamics by setting fixed poverty lines at 25% and 40% as anchor points, and subsequently counted households to see whether more or less fall below these anchor points at certain places or points in time . Hence, their approach is more rigid, but allows for direct temporal and regional comparison.
The focus on relative wealth dynamics seems appropriate as people tend to evaluate their living conditions by comparing their current with their previous circumstances or with the circumstances of people in their surroundings. This fact may also explain the findings of Goodhand (2003) that "absolute measures of poverty may be less significant (...) as triggers to violence" and that "transient poverty is likely to have a more significant influence on the dynamics of war and peace than chronic poverty" .
However, given the analytical approach, which is self-contained and not open as opposed to income or consumption studies, households have limited options for mobility. Only households in intermediate quintiles can truly rise and fall in the frame set by asset scores and wealth quintiles. In general, it is possible that asset-based wealth indices are more stable than other SEP indicators (e.g. absolute poverty in monetary terms, income or consumption). In times of economic hardship, households may draw upon other resources before selling their assets. Likewise, in prosperous periods, it may take some time until economic success is reflected in asset possession and an associated wealth index.
As substantial negative consequences of the armed conflict in Côte d'Ivoire are demonstrated in other micro- and macro-level reports, and even by some of our own indicators, it is likely that our PCA-based analysis failed to detect at least some of the socioeconomic dynamics and associated predictors. Hence, it could be worthwhile to use a more comprehensive livelihood framework to assess the impact of armed conflicts in future investigations (see endnote 8). Furthermore, it would be useful to compare our results with other war-torn as well as peaceful settings in order to get a better idea about the threats and opportunities imposed on households by armed conflict as well as to further verify causality. In fact, high-quality data for comparison might be readily available from a growing number of demographic surveillance systems (DSS; see http://www.indepth-network.org; accessed 23 August 2010) or the comparatively new Household in Conflict Network (see http://www.hicn.org; accessed 23 August 2010).
In conclusion, we emphasize that more micro-level research on the measurement of SEP in low-income countries as well as on the impact of armed conflict and war is warranted. In general, methods to assess and meaningfully interpret longitudinal micro-level wealth data should be further developed. Recent methodological reviews demonstrated that all actual approaches for measuring SEP in low-income countries have their drawbacks. Furthermore, the results of the different methods showed only limited agreement and have restrictions for further processing [59, 61–64]. Along with previous research undertaken in conflict zones elsewhere in Africa [37, 65–76], the present study can be considered as additional evidence of the feasibility of research even in troubled times and zones. Knowledge about the profiles of households that are more resilient to armed conflict could help to better prevent and/or alleviate adverse conflict-related and increasingly civilian-borne socioeconomic effects. Consequently, we would like to reinforce Tam and colleagues' "call to arms"  in order to boost research related to armed conflict and war.