Jordan - Labor Market Panel Survey, JLMPS 2016
"As the second wave of the JLMPS longitudinal study, the JLMPS 2016 both followed the 2010 panel and added a refresher sample. For the panel component of the data, we attempted to recontact all households that were included in the 2010 wave. Among the households that were found, we also followed any split households. Split households occur when one or more individuals from 2010 leave their 2010 household to form a new household. For example, an individual who was the son of the household head in 2010 might marry and form a new household. The entire new household is included in our sample, including members who were not part of the 2010 sample. The refresher sample over-sampled neighborhoods in Jordan that, as of the 2015 Census, had a high proportion of non-Jordanians. The final JLMPS 2016 sample is made up of 7,229 households, including 3,058 that were part of the original 2010 sample, 1,221 split households and 2,950 refresher households. The JLMPS 2016 sample captured a total of 33,450 individuals. We discuss the sampling strategy and the creation of the sampling and attrition weights in detail below.
----> 2010 sample
The 2010 sample was a nationally-representative sample designed to represent urban and rural areas in the three regions of Jordan: North, Middle, and South. For sampling purposes, the sample was stratified into 30 strata based on a combination of the 12 governorates of Jordan and five different location classifications within them: (1) basic urban (2) rural (3) large central city urban in Amman, Zarqa, and Irbid governorates (4) suburban Amman and Zarqa and (5) exurban Amman. The 2010 sample captured 5,102 households and 25,953 individuals.
----> Refresher sample
The refresher sample in 2016 was designed to over-sample neighborhoods with high proportions of non-Jordanians. The prior, 2010 wave, was just before the Arab Spring and subsequent conflicts in the region. Although Jordan itself did not have internal conflict, its neighbors of Iraq and Syria did. Jordan now hosts a large number of refugees from these conflicts. Based on the Jordanian Census of 2015, there were 9.5 million individuals in Jordan, of whom 6.6 million were Jordanian and 1.3 million were Syrian (Department of Statistics. Jordan also hosts a large population of migrant workers, including 636,000 Egyptians as of 2015. UNHCR's estimate of the number of registered Syrian refugees in Jordan as of September 2017 was 654,000. Jordan also hosts a number of Palestinians, with substantial waves of arrivals around 1948 and 1967. Individuals of Palestinian origin are mostly naturalized and therefore counted in the Jordanian population. However, non-nationalized Palestinians were the third largest group after Syrians and Egyptians in Jordan, at around 634,000 individuals in 2015. There were also around 131,000 Iraqis and smaller populations from numerous other countries. Altogether, these non-Jordanians play a large and increasing role in the Jordanian economy. The refresher sample was designed to over-sample these groups in order to ensure national representativeness in the JLMPS 2016, as well as sufficient observations for analysis of different groups, such as Syrian refugees.
The sampling frame for the refresher sample was Jordan's 2015 Population and Housing Census. The census was fielded in late November of 2015. There were 6.6 million Jordanians in 1.4 million households, including Palestinians with Jordanian citizenship. There were 1.3 million Syrians in 0.2 million households at the time of the Census. There were 0.6 million Egyptians and 0.8 million Other Arabs. Other Arabs are primarily Palestinians who are not citizens, historically from Gaza, more recently some are Palestinians from Syria. There are also 0.2 million individuals of other nationalities. In total, there were 1.9 million households and 9.5 million individuals. These census data (geographically disaggregated, as discussed below) are also the source of our expansion factors for the JLMPS weights.
In order to over-sample areas with high proportions of non-Jordanians, we examined the distribution of households with non-Jordanian heads (hereafter referred to as non-Jordanian households). Our goal was to create two strata, one with a high proportion of non-Jordanian households and one with a low proportion of non-Jordanian households in order to oversample the former. The information on household nationality was assessed at the lowest geographical level possible, the neighborhood (hayy). This is the cluster or primary sampling unit (PSU) level we used for drawing our refresher sample. To draw the sample, we identified the 90th percentile of neighborhoods as having 45.7% non-Jordanian households. Thus, 45.7% and higher shares of non-Jordanian household are our "high" non-Jordanian strata and shares lower than 45.7% are our "low" non-Jordanian strata. We further stratified our refresher sample along two dimensions: governorate and location (urban, rural, or refugee camps). The camps were the two official camps in Jordan: Zaatari refugee camp, in the Mafraq governorate, and Azraq refugee camp, in the Zarqa governorate.
The high non-Jordanian and camps strata were both over-sampled in order to provide a sufficient number of observations for research and analysis. This over-sampling strategy is accounted for in our weights, discussed below. Across the strata, a total of 200 PSUs (neighborhoods) were selected. Within each PSU, the plan was to sample 15 households.
----> Sample attrition
For the panel data, tracking households from 2010 to 2016, a key issue is sample attrition. There are two points in time when attrition can occur: between the 2010 round and 2016 enumeration and between 2016 enumeration and 2016 fielding. There are also two types of attrition that can occur: Type I attrition occurs when we cannot locate a 2010 household at all, while Type II attrition occurs when we can locate a 2010 household, it has a split, and we cannot locate the split household. This section discusses the patterns of the two different types of attrition and then presents the models predicting attrition that are used as inputs into generating the sample weights.
-----> Attrition of entire households (Type I attrition)
In undertaking the enumeration and fieldwork, a key goal was to relocate as many 2010 households as possible. At the enumeration stage, from the original 2010 sample of 5,102 households, 3,427 were re-located. In the cases when households were not located, if possible, data were collected on the status of the household or the reason they were not present. During enumeration, there were 81 households that had left the country entirely (all members left) and 44 households that had all died (all members died). We refer to these cases of all the members leaving or dying as "natural attrition." We do not include cases of natural attrition in our calculation of attrition rates or our attrition models, as these households do not exist (in our sampling frame) in 2016.
At the enumeration stage, we were unable to locate 1,481 households and 69 households refused (both these results are forms of attrition). Thus, our Type I attrition rate was 31.1% at the enumeration stage. After updates during fielding, from the 3,427 households found during enumeration, 26 households left the country, 8 died out, 179 could not be found, and 157 refused. Thus, our final Type I attrition rate was 38.2%. Of the 5,102 households from 2010, 3,057 remained in the sample.
-----> Attrition of split households (Type II attrition)
One of the lessons we learned from ELMPS 2012 was that we need to account for attrition between enumeration and fielding on the individual level as well as the household level. We therefore included essentially the same questions as from enumeration in order to update the disposition of different individuals who were in the 2010 wave and present at enumeration. This also allowed us to track additional split households that occurred between enumeration and fielding. Unfortunately, the additional split households were not followed up in the field. However, we can use the data on individuals who died, left the country, or moved to group housing, thus leaving the sample frame, between enumeration and fielding to assess natural attrition as distinct from Type II attrition. Split households between enumeration and fielding thus contribute to Type II attrition.
The status of individuals is only shown for those whose 2010 household was found. The households found in enumeration, during 2010, contained 18,227 individuals. Of these, 15,617 were still present in their original households. Among those no longer present, four had moved to group housing, 234 emigrated, and 382 died, totaling 620 individuals who left the sample due to natural attrition. The remaining 1,990 individuals formed split households. Since individuals can split together, we identify individuals who form a new household together as one "split household." There were 1,911 split households at enumeration, of which 1,536 were found, for a Type II attrition rate of 19.6%. Since additional households were lost between enumeration and fielding, there were only 15,357 individuals from 2010 who could potentially be in their original (or split) household at the fielding stage. We successfully located 14,502 of these individuals from 2010 during fielding. Of the 855 individuals lost, 208 were lost to natural attrition, and 647 were lost into 616 split households. When looking at the final status of individuals, there were 16,631 individuals who were present in 2010 in the households that were successfully found at fielding. Of these, 13,235 individuals were found in their original households. Of the remaining 3,396, in total 757 were lost to natural attrition. There were 2,639 individuals who split, into 2,465 split households. Half of the split households were found, for a Type II attrition rate of 50.5%. (Krafft and Assaad, 2018)"
For more information on the survey sampling procedure, see the paper cited in the "Citations" section.
Deviations from Sample Design
"There were a few deviations from planned sampling in implementation. First, in the high strata an additional PSU in urban Amman and an additional PSU in rural Amman were added. Ajloun and Tafileh high strata rural areas were not sampled. One less PSU was drawn from the high strata rural area in Mafraq. In total, 199 PSUs were sampled. There were also some deviations from the planned number of households in each PSU. The goal was to sample 15 households per PSU, and for 93.5% of PSUs, this was achieved. Most other PSUs (3.5%) sampled 14 households. Two PSUs sampled only 13 households, one PSU only 12 households, two PSUs 11 households, and one PSU only 2 households. In terms of an overall response rate, the mean response rate was 98.8%. Within the different dimensions of the strata, missing households were slightly more common in rural areas: a 97.8% response rate in rural areas, a 98.9% response rate in urban areas, and a 100% response rate in the camps. These response rates on the PSU level are factored in to our sample weights, as discussed below. Ultimately, of a planned 3,000 households, our refresher sample contained 2,950 households with 13,423 individuals" (Krafft and Assaad, 2018).
----> Models of sample attrition
"We model sample attrition for two reasons; first, we want to examine whether attrition is random or related to household characteristics. Second, if there are differences in attrition related to observable characteristics, we want to incorporate those differences into our weighting strategy for the panel households. Households that naturally attrited are excluded from the model. Characteristics are, necessarily, from the 2010 data.
There are some significant predictors of attrition. In terms of household composition, households with more working age and especially more elderly (65+) females are significantly less likely to attrite. Households composed of all males, compared to mix sex households, were significantly more likely to attrite. There were not significant differences by the geographic strata that were used in 2010 for stratifying the sample, which is encouraging for sample representativeness. Households in the top wealth decile were significantly more likely to attrite than the poorest, but there were not differences for other quintiles and this difference was driven by urban areas; although the rural and wealth decile interactions were not significant, they show lower odds (essentially canceling out the higher odds of the main effect).
In terms of governorates, there are significantly lower odds of attrition for Karak, Tafileh, and Ma'an, but higher odds of attrition in Aqaba (all in the South region), compared to Amman. There are also significant interactions with rural, lower odds of attrition, for Zarqa, Irbid, and Aqaba (the latter the opposite of the main effect), and significantly higher odds in rural Tafileh. Homeownership predicts significantly lower attrition. There are not significant differences by head age group or sex, although households headed by females 25-34 were significantly more likely to attrite. There were not significant differences by marital status, but there was a significantly higher probability of attrition for households with divorced, female heads.
Households with more educated heads were more likely to attrite, significantly so for secondary and higher education, as compared to less than basic. There were few significant head labor market characteristics, which bodes well for the labor market representativeness of our panel. Unemployed individuals were significantly less likely to attrite, while those out of the manpower basis (disabled or elderly) were more likely to attrite than the reference public sector. There were no significant rural and labor market status interactions. Overall, the model had a pseudo R-squared of 14.7%
For the Type II attrition model, the sample is restricted to those 2,386 splits with heads age 6+ (who have individual characteristics from 2010). Since most households include few members, we only model composition in terms of additional working age males and females. There are significantly lower odds of attrition for an additional working age male. Here, there are significantly higher odds of attrition for strata other than the reference urban (not large city). Splits from rural areas and the exurbs in particular have significantly higher odds or attrition. There are significantly higher odds of attrition for a number of wealth quintiles compared to the poorest, although essentially the differences imply that splits from the poorest quintile are less likely to attrite. There are some significantly lower odds of attrition for wealth rural interactions. As was the case for Type I attrition, it appears to be primarily urban and wealthier households driving differences in attrition. Only splits from Madaba have a significantly different (higher) odds of attrition. There are no significant rural and governorate interactions. Splits from households that owned their home are significantly less likely to attrite. Compared to split heads <25, only those 45+ are significantly more likely to attrite. The odds ratio here is high; one possible reason is that these are splits where the split head died or moved to group housing, but that this was not captured in the field.
Female-headed splits are significantly less likely to attrite, although the female and 25-34 interaction is a significantly higher odds of attrition. Compared to those splits with less than basic, other categories are less likely to attrite, all but higher ed. significantly so. This may be related to communication around new location. Compared to splits whose head, in 2010, was a public sector worker, formal private wage, informal private wage, and out of labor force individuals are more likely to attrite. Overall, the pseudo R-squared of the Type II attrition model is 10.0%" (Krafft and Assaad, 2018).
For more information on the sampling weights, see the paper cited in the "Citations" section.