Nigeria - COVID-19 National Longitudinal Phone Survey 2020
Reference ID | NGA-2020-NLPS-v07-M |
Year | 2020 |
Country | Nigeria |
Producer(s) | National Bureau of Statistics (NBS) - Federal Government of Nigeria |
Sponsor(s) | Bill and Melinda Gates Foundation - BMGF - Funded the study Federal Government of Nigeria - FGN - Funded the study United States Agency for International Development - USAID - Funded the study |
Metadata | Documentation in PDF Download DDI Download RDF |
Created on | Mar 17, 2021 |
Last modified | Mar 17, 2021 |
Page views | 967793 |
Downloads | 8864 |
Sampling
Sampling Procedure
Wave 4 of the GHS-Panel conducted in 2018/19 served as the frame for the Nigeria COVID-19 NLPS survey. The GHS-Panel sample includes 4,976 households that were interviewed in the post-harvest visit of the fourth wave in January/February 2019. This sample of households is representative nationally as well as across the 6 geopolitical Zones that divide up the country. In every visit of the GHS-Panel, phone numbers are collected from interviewed households for up to 4 household members and 2 reference persons who are in close contact with the household in order to assist in locating and interviewing households who may have moved in subsequent waves of the survey. This comprehensive set of phone numbers as well as the already well-established relationship between NBS and the GHS-Panel households made this an ideal frame from which to conduct the COVID-19 monitoring survey in Nigeria.
Among the 4,976 households interviewed in the post-harvest visit of the GHS-Panel in 2019, 4,934 (99.2%) provided at least one phone number. Around 90 percent of these households provided a phone number for at least one household member while the remaining 10 percent only provided a phone number for a reference person. Households with only the phone number of a reference person were expected to be more difficult to reach but were nonetheless included in the frame and deemed eligible for selection for the Nigeria COVID-19 NLPS.
To obtain a nationally representative sample for the Nigeria COVID-19 NLPS, a sample size of approximately 1,800 successfully interviewed households was targeted. However, to reach that target, a larger pool of households needed to be selected from the frame due to non-contact and non-response common for telephone surveys. Drawing from prior telephone surveys in Nigeria, a final contact plus response rate of 60% was assumed, implying that the required sample households to contact in order to reach the target is 3,000.
3,000 households were selected from the frame of 4,934 households with contact details. Given the large amount of auxiliary information available in the GHS-Panel for these households, a balanced sampling approach (using the cube method) was adopted. The balanced sampling approach enables selection of a random sample that still retains the properties of the frame across selected covariates. Balancing on these variables results in a reduction of the variance of the resulting estimates, assuming that the chosen covariates are correlated with the target variable. Calibration to the balancing variables after the data collection further reduces this variance (Tille, 2006). The sample was balanced across several important dimensions: state, sector (urban/rural), household size, per capita consumption expenditure, household head sex and education, and household ownership of a mobile phone.
Response Rate
BASELINE (ROUND 1): All 3,000 households were contacted in the baseline round of the phone survey. 69 percent of sampled households were successfully contacted. Of those contacted, 94 percent or 1,950 households were fully interviewed. These 1,950 households constitute the final successful sample and will be contacted in subsequent rounds of the survey.
ROUND 2: Interviewers attempted to contact and interview all 1,950 households that were successfully interviewed in the baseline of the COVID-19 NLPS. 1,852 households (95% of the 1,950 attempted) were contacted and 1,820 (93.3%) were successfully interviewed in the second round. Of those contacted, 22 households refused outright to be interviewed and 10 were partially interviewed.
ROUND 3: Interviewers attempted to contact and interview all 1,925 households that were successfully interviewed in the Baseline of the COVID-19 NLPS, excluding 25 households that had refused in Round 2. Thus, the sample included households that were not successfully interviewed in Round 2, in an effort to maintain the sample size. 1,837 households (95.4% of the 1,925 attempted) were contacted and 1,790 (93%) were successfully interviewed in the third round. Of those contacted, 28 households refused outright to be interviewed and 18 were partially interviewed. Of the 1,790 successfully interviewed households, 1,737 were households that have been successfully interviewed in all three rounds of the survey so far. These are the households that form a complete panel across the three rounds.
ROUND 4: Interviewers attempted to contact and interview all 1,881 households that were successfully interviewed in the Baseline of the COVID-19 NLPS, excluding 69 households that had refused in Round 2 or Round 3. Thus, the sample included households that were not successfully interviewed subsequent to the baseline in an effort to maintain the sample size. As shown in 7-11, 1,819 households (96.7% of the 1,881 attempted) were contacted and 1,789 (95.1%) were successfully interviewed in the fourth round. Of those contacted, 19 households refused outright to be interviewed and 9 were partially interviewed. Of the 1,789 successfully interviewed households, 1,691 were households that have been successfully interviewed in all four rounds of the survey so far. These are the households that form a complete panel across the four rounds.
ROUND 5: Interviewers attempted to contact and interview all 1,856 households that were successfully interviewed in the Baseline of the COVID-19 NLPS, excluding 94 households that had refused in previous rounds of the survey Thus, the sample included households that were not successfully interviewed subsequent to the baseline in an effort to maintain the sample size. As shown in Table 7-15 of the BID, 1,794 households (96.7% of the 1,856 attempted) were contacted and 1,774 (95.6%) were successfully interviewed in the fifth round. Of those contacted, 13 households refused outright to be interviewed and 5were partially interviewed. Of the 1,774 successfully interviewed households, 1,656 were households that have been successfully interviewed in all five rounds of the survey so far. These are the households that form a complete panel across the five rounds.
ROUND 6: Interviewers attempted to contact and interview all 1,839 households that were successfully interviewed in the Baseline of the COVID-19 NLPS, excluding 111 households that had refused in previous rounds of the survey. Thus, the sample included households that were not successfully interviewed subsequent to the baseline in an effort to maintain the sample size. As shown in Table 7-19 of the BID, 1,781 households (96.8% of the 1,839 attempted) were contacted and 1,762 (95.8%) were successfully interviewed in the sixth round. Of those contacted, 8 households refused outright to be interviewed and 11 were partially interviewed. Of the 1,762 successfully interviewed households, 1,640 were households that have been successfully interviewed in all six rounds of the survey so far. These are the households that form a complete panel across the six rounds.
ROUND 7: Interviewers attempted to contact and interview all 1,811 households that were successfully interviewed in the Baseline of the COVID-19 NLPS, excluding 139 households that had refused in previous rounds of the survey. Thus, the sample included households that were not successfully interviewed subsequent to the baseline in an effort to maintain the sample size. As shown in Table 7-23 of the BID, 1,740 households (96.8% of the 1,811 attempted) were contacted and 1,726 (95.3%) were successfully interviewed in the seventh round. Of those contacted, 4 households refused outright to be interviewed and 10 were partially interviewed. Of the 1,726 successfully interviewed households, 1,573 were households that have been successfully interviewed in all seventh rounds of the survey so far. These are the households that form a complete panel across the seventh rounds.
ROUND 8: Interviewers attempted to contact and interview all 1,810 households that were successfully interviewed in the Baseline of the COVID-19 NLPS, excluding 139 households that had refused in previous rounds of the survey. Thus, the sample included households that were not successfully interviewed subsequent to the baseline in an effort to maintain the sample size. As shown in Table 7-27 of the BID, 1,738 households (96.0% of the 1,810 attempted) were contacted and 1,723 (95.2%) were successfully interviewed in the eighth round. Of those contacted, 9 households refused outright to be interviewed and 6 were partially interviewed. Of the 1,723 successfully interviewed households, 1,547 were households that have been successfully interviewed in all eight rounds of the survey so far. These are the households that form a complete panel across the eight rounds.
Weighting
In order to produce national estimates from the successfully interviewed sample, weights must be applied to the information provided by sampled households. Weights for the GHS-Panel serve as the basis for the Nigeria COVID-19 NLPS, but the weights must be adjusted to reflect the selection and interviewing process. The weights for the Nigeria COVID-19 NLPS were calculated in several stages.
1. Begin with the GHS-Panel full sample household weights.
2. Apply an adjustment factor for the selection into the frame (GHS-Panel households that have contact details). A ratio adjustment was applied at the Zone-level (the strata for the GHS-Panel) to preserve the sum of household weights within each Zone between the full GHS-Panel sample and the NLPS frame.
3. Apply an adjustment for selection into the NLPS sample. The adjustment is a simple expansion factor that is the inverse of the selection probability from the frame for each sampled unit.
4. Apply an adjustment factor for non-contact of sampled households. This was again performed with a ratio adjustment at the Zone-level.
5. Apply an adjustment factor for non-response of contacted households through a ratio adjustment at the Zone-level.
6. Calibrate the weights (following adjustments 2-5) according to the properties of the full weighted GHS-Panel sample. This calibration step adjusts the weights such that the estimates obtained from the final NLPS sample will match the weighted means of the full GHS-Panel sample for specified characteristics. The calibration was performed using only information obtained from the GHS-Panel interview and thus will only reflect changes in the sample composition and not changes over time. The calibration applied here aims to correct for selection bias that is introduced at any point between identification of the frame and the final successfully interviewed sample. Selection bias is of particular concern in phone surveys since some segment of the population does not have access to a phone and there are more difficult barriers to successfully reach and interview households over the phone. The calibration was applied using the ReGenesees package in R. The characteristics that were considered in the calibration were the same factors included in the balanced sample selection described in 3.1 above. The weights were also applied to the total number of households in the population given by the GHS-Panel weights.
7. Trim the weights. Outlier weights were trimmed using the ReGenesees package in R which adjusts the weights to given bounds while minimizing the deviation from the estimates obtained from the calibration in step 6.
In subsequent rounds of the survey, steps 4, 5, and 6 will be applied to the final baseline weights.
BASELINE (ROUND 1): The weights can be found in the household-level data file (r1_sect_a_3_4_5_6_8_9_12). The variable name is wt_baseline.
ROUND 2: The baseline weights were adjusted for noncontact and nonresponse as well as calibrated following the same procedures outlined above (steps 4, 5 and 6). The round 2 weights can be found in the household-level data file (r2_sect_a_2_5_6_8_12). The variable name is wt_round2.
ROUND 3: In Round 3, two different weights are provided: cross section and panel weights. The cross section weights are applicable to the entire round 3 sample while the panel weights are only applicable to round 3 sample households that have been successfully interviewed in all three rounds of the survey so far. For both of these weights, the baseline weights were adjusted for noncontact and nonresponse as well as calibrated following the same procedures outlined above (steps 4, 5 and 6). The round 3 weights can be found in the household-level data file (r3_sect_a_2_5_6_12). The cross section weight is contained in the variable named wt_round3 while the panel weight is contained in the variable named wt_r3panel.
ROUND 4: In Round 4, two different weights are provided: cross section and panel weights. The cross section weights are applicable to the entire round 4 sample while the panel weights are only applicable to round 4 sample households that have been successfully interviewed in all four rounds of the survey so far. For both of these weights, the baseline weights were adjusted for noncontact and nonresponse as well as calibrated following the same procedures outlined above (steps 4, 5 and 6). The round 4 weights can be found in the household-level data file (r4_sect_a_2_5_5b_6_8_9_12). The cross section weight is contained in the variable named wt_round4 while the panel weight is contained in the variable named wt_r4panel.
ROUND 5: In Round 5, several different weights are provided: two at the household-level and three at the individual-level. The two household weights are the same as have been provided in previous rounds, that is cross section and panel weights. The cross section weights are applicable to the entire round 5 sample while the panel weights are only applicable to round 5 sample households that have been successfully interviewed in all five rounds of the survey so far. For both of these weights, the baseline weights were adjusted for noncontact and nonresponse as well as calibrated following the same procedures outlined in section 2.2 (steps 4, 5 and 6). The round 5 household weights can be found in the household-level data file (r5_sect_a_2_5c_6_12). The cross-section weight is contained in the variable named wt_round5 while the household panel weight is contained in the variable named wt_r5panel.
Given the focus on individual employment information in round 5 and the selection steps outlined above for the sample of working age members, an additional three individual-level weights were calculated and provided in the round 5 data. The individual weights for the employment module were calculated according to:
w_ish=w_h×(n_hs/N_hs )^(-1)
Where w_ih is the sampling weight for individual i who is sex s (male or female) in household h, w_h is the final household level weight (i.e. wt_round5), N_hs is the total number of eligible household members (aged 15-64) of sex s in household h and n_hs is the equivalent number of selected eligible individuals in the household. The individual weights were then calibrated to correspond to the sex and age distribution of the total working age population according to the post-harvest visit of the GHS-Panel.
The basic individual weight described above is the cross section individual weight that considers all individuals that employment information was collected on. This weight is called wt_employ_r5 and can be found in the individual-level employment data file (r5_sect_6b). However, an additional two weights are provided for the panel of individuals interviewed in the GHS-Panel wave 4 and round 5 of the NLPS (i.e. excluding individuals added in the NLPS). The first weight (wt_employ_r5_pp_panel) contains the weight for individuals interviewed in the post-planting visit of the GHS-Panel wave 4 and the second (wt_employ_r5_ph_panel) contains the weight for individuals interviewed in the post-harvest visit of the GHS-Panel wave 4.
ROUND 6: In Round 6, several different weights are provided: two at the household-level and three at the individual-level. The two household weights are the same as have been provided in previous rounds, that is cross section and panel weights. The cross section weights are applicable to the entire round 6 sample while the panel weights are only applicable to round 6 sample households that have been successfully interviewed in all six rounds of the survey so far. For both of these weights, the baseline weights were adjusted for noncontact and nonresponse as well ascalibrated following the same procedures outlined in section 2.2 (steps 4, 5 and 6). The round 6 household weights can be found in the household-level data file (r6_sect_a_2_3a_6_9a_12). The cross section weight is contained in the variable named wt_round6 while the household panel weight is contained in the variable named wt_r6panel.
Given the focus on individual education information in round 6 and the selection steps outlined above for the sample of school-aged members (5-18 years), an additional three individual-level weights were calculated and provided in the round 6 data. The individual weights for the employment module were calculated according to:
w_ish=w_h×(n_hs/N_hs )^(-1)
Where w_ih is the sampling weight for individual i who is sex s (male or female) in household h, w_h is the final household level weight (i.e. wt_round6), N_hs is the total number of eligible household members (aged 5-18) of sex s in household h and n_hs is the equivalent number of selected eligible individuals in the household. The individual weights were then calibrated to correspond to the sex and age distribution of the total school-aged population according to the post-harvest visit of the GHS-Panel.
The basic individual weight described above is the cross section individual weight that considers all individuals that employment information was collected on. This weight is called wt_educ_r6 and can be found in the individual-level education data file (r6_sect_5c.dta). However, an additional two weights are provided for the panel of individuals interviewed in the GHS-Panel wave 4 and round 6 of the NLPS (i.e. excluding individuals added in the NLPS). The first weight (wt_educ_r6_pp_panel) contains the weight for individuals interviewed in the post-planting visit of the GHS-Panel wave 4 and the second (wt_educ_r6_ph_panel) contains the weight for individuals interviewed in the post-harvest visit of the GHS-Panel wave 4.
ROUND 7: In Round 7, two different weights are provided: cross section and panel weights. The cross section weights are applicable to the entire round 7 sample while the panel weights are only applicable to round 7 sample households that have been successfully interviewed in all four rounds of the survey so far. For both of these weights, the baseline weights were adjusted for noncontact and nonresponse as well as calibrated following the same procedures outlined in section 2.2 (steps 4, 5 and 6). The round 7 weights can be found in the household-level data file (r7_sect_a_5_6_8_9_12). The cross section weight is contained in the variable named wt_round7 while the panel weight is contained in the variable named wt_r7panel.
ROUND 8: In Round 8, two different weights are provided: cross section and panel weights. The cross-section weights are applicable to the entire round 8 sample while the panel weights are only applicable to round 8 sample households that have been successfully interviewed in all eight rounds of the survey so far. For both of these weights, the baseline weights were adjusted for noncontact and nonresponse as well as calibrated following the same procedures outlined in section 2.2 (steps 4, 5 and 6). The round 8 weights can be found in the household-level data file (r8_sect_a_2_5c_6_12). The cross section weight is contained in the variable named wt_round8 while the panel weight is contained in the variable named wt_r8panel.