SAMPLE DESIGN APPENDIX A

Transcription

1 SAMPLE DESIGN APPENDIX A A.1 SAMPLE SIZE AND ALLOCATION The Armenia Demographic and Health Survey (ADHS) required a nationally representative sample o women age and men age The sample was designed to provide estimates o most survey indicators (including ertility, abortion and contraceptive prevalence) or Armenia as a whole, or three residence categories (Yerevan, other urban and rural areas) and or each o ten administrative regions (marz). The design also called or estimates o inant and child mortality at the national level and or the three residence categories (Yerevan, other urban and rural areas). The target sample size o 6,500 completed interviews with women in the childbearing ages was allocated as ollows: 1,500 to Yerevan, and 500 to each o the ten regions. Within each region, the sample was allocated between urban and rural strata in proportion to the population size. This yielded 1 sample strata. Table A.1. Overall, the sample allocation resulted in 1,500 emale respondents in Yerevan,,300 in other urban areas and,700 in the rural areas. A two-stage sample design was used. The irst stage selected 60 area units (i.e., sample clusters) rom the sampling rame provided by the National Statistical Service. The second stage selected households in which all women were eligible to be interviewed. The sample was developed to yield, on average, 5 emale respondents rom each sample cluster. Additional description o the sampling rame and the two stages o selection are provided below. Interviews were completed with 6,430 women. Men age were interviewed in every third household; this yielded 1,719 completed interviews. A. AREA FRAME The rame consisted o the list o the area units in the 1996 Data Base o Addresses and Households, a household listing carried out by the National Statistical Service in 1996 covering the whole country. There were a total o 1,03 areas demarcated in the rame. Except or the two largest cities, which were divided into sectors, each area listed in the rame corresponded to a whole town or village. The rame provided identiication inormation or each region, subregion (i applicable), and locality, as well as urban-rural classiication and the altitude o the area (classiied into three categories: <1,300 meters, 1,300-1,700 meters and >1,700 meters). The measures o size in urban areas were the 1996 population counts o individual areas. In rural areas, the measures o size were deined as the number o households in the village, multiplied by the average household size in the rural part o the region in which the village was located. The reason or this decision was some uncertainty in the population igures or individual villages, while the inormation on the number o households appeared more reliable. Note that, when summed over the rural sector o a region, the total rural measure o size remained equal to the total population count o the 1966 Data Base, so that the igures in Table 1 were not aected. Appendix A 17

2 Table A.1 Sample allocation by region and by residence Number o Number o primary Region/ households Population Target sampling residence Residence (1996) (1996) sample units Region Yerevan 1 5,840 1,036, Aragatsotn 1 9,918 44, , , Ararat 1 18,981 81, ,55 04, Armavir 1, , , , Gegharkunik 1,937 9, , , Lori 1 53,578 01, , 118, Kotayk 1 41,31 17, , , Shirak 1 61,399 33, , , Syunik 1 5, , ,415 48, Vayots Dzor 1 6,179 6, ,537 40, Tavush 1 14,015 57, ,761 9, Residence Yerevan 5,840 1,036,79 1, Other urban 76,964 1,116,00,31 96 Rural 345,008 1,199,381, Armenia 874,81 3,351,680 6, = Urban = Rural 18 Appendix A

3 For the selection o the sample, areas were arranged according to the ollowing ive variables, in the order speciied: 1. Region (i.e. marz) (00-10). Urban-rural (1-) 3. Altitude (1-3) 4. Subregion (where speciied) 5. Population (i.e. measure o size) o the area. The irst two variables were used to deine the explicit strata or the purpose o selection (i.e. or each region a pre-speciied number o urban and rural primary sampling units (PSUs) were selected independently. The remaining three variables provided implicit ordering o the list or systematic selection. A.3 SELECTION OF PRIMARY SAMPLING UNITS The initial phase o the selection o PSUs required two steps: irst the selection o area units rom the 1,03 areas in the sampling rame by systematic sampling with probability proportional to size (PPS). A total o 11 areas were selected. However, 5 o the selected areas were particular large (i.e., sel-representing and were selected more than once by the systematic PPS sampling), so it was necessary to select more than one PSU rom those 5 areas. From those 5 areas, a total o 74 PSUs were created. Overall, these 74 PSUs and the 186 (11-5) non sel-representing PSUs provided a total o 60 PSUs. At this point, the overall sampling probability or each region ( ) and an initial irst stage sampling probability or each selected PSU ( 1 ) were known. A.4 SECONDARY SAMPLING UNITS AND SEGMENTATION From the perspective o cost and the availability o resources, most o the 60 PSUs were too large to perorm a complete household listing operation. Accordingly almost all PSUs were subdivided into a pre-speciied number a o secondary sampling units (SSUs). The creation o SSUs, when possible, was done in the oice based on the boundaries and landmarks shown in the mapping materials rom the 1996 database. The created SSUs had clearly identiiable boundaries and a known measure o size. One o the created SSUs was selected with PPS, i.e. with probability M a i p i = with M = Σ M M i= 1 i where M i is the measure o size or the i th SSU. Appendix A 19

4 The task o household listing was urther reduced by segmentation. Each SSU was divided into 8 segments. The segment boundaries were identiied in the ield. The segments were grouped to orm 4 pairs, grouping the largest segment with the smallest, the next largest with the next smallest, etc. Measure o size, s i, or every pair o segments was obtained either rom the 1996 Data Base or rom a quick count done in the ield. Let s i be the measure o population size or pair within SSUi. One o the pairs was selected with PPS, i.e. with probability p i = s i / s i with s i = Σ i s i Ater segmentation, the irst stage sampling probability o the selected PSUs was: ' 1 = p i. p i. 1, (p i. p i < 1) A.5 HOUSEHOLD LISTING AND SELECTION A complete household listing was conducted within the selected pairs o segments in order to construct the sampling rame or the selection o households. The required household stage sampling rate was: = ( ) /( 1' ) In all PSUs the sampling rate or the selection o households within listed segments was close to one in eight (i.e., 0.15). This outcome was by design. The number o SSUs created in each PSU was set to obtain this result. A relatively similar sampling rate across sample segments meant a variable take rom each although, on average, the target number o completed interviews with emale respondents remained 5 per PSU. A.6 ADJUSTMENT OF HOUSEHOLD SAMPLING RATE AFTER LISTING Since there was some doubt about the population size measures in the 1996 database, the second stage sampling rates were adusted so as to control the inal sample size. This was an overall adustment, the same or the whole sample, so as not to aect the planned relative sampling rates. For each PSU (sample area k in domain ), let: L = the number o households listed in the selected PSU (the selected pair o segments as deined above or the whole area i not segmented). ' = the second stage sampling raction or the household selection in a PSU. This means that the number o households expected to be selected is h = L. ' 0 Appendix A

5 Let X be the conversion actor rom households to completed women interviews in domain : X = H. W. r where H is the average household size, W is the proportion o the population who are women age (i.e., the expected number o eligible women per person in the population) and r the expected response rate in the domain. Overall country-level igures were used: H =3.84, W =0.78 and r =0.94 (6 percent non-response). This gives the expected number o completed interviews as n = Σ [ X Σ ( L. ' )]. k summed over all PSUs (segments or localities) in the sample. To achieve the required sample size n=6,500 completed interviews, the second stage sampling ractions was adusted throughout by the actor ( n n = 6500 /5403 = 1.0), i.e. modiied in each area as n ' ' =. '. n On the basis o the inal sampling ractions, households were selected systematically rom geographically ordered household listings. A.7 RESPONSE RATES Tables A.1 and A. present detailed inormation on the results o the household and individual interviews. Household interviews were completed or 97 percent o the occupied households. A total o 6,685 eligible women rom these households and 1,913 eligible men rom every third household were identiied or the individual interviews. O the eligible women identiied, 96 percent were successully interviewed; o the eligible men, 90 percent were successully interviewed. The principal reason or non-response among eligible women and men was the ailure to ind them at home despite repeated visits to the household. The reusal rate was low. There is no dierence by urban-rural residence in the response rates or eligible women and men. Appendix A 1

6 Table A. Sample Implementation: women Percent distribution o households and eligible women in the DHS sample by result o the interview and household, eligible women and overall response rates, according to region and urban rural Area, Armenia 000 Result o interview and response rate Urban Rural Total Household interviews Completed (C) No competent respondent (HP) Reused (R) Dwelling not ound (DNF) Absent (A) Dwelling vacant (DV) Total percent Number 3,69,895 6,54 Response rate (HRR) Women interviews Completed (EWC) Not at home (EWNH) Reused (EWR) Partly completed (EWPC) Incapacitated (EWI) Total percent (EWO) Number 3,699,986 6,685 Response rate (EWRR Overall response rate (ORR) Using the number o households alling into speciic response categories, the household response rate (HRR) is calculated as: 100 x C C + HP + R + DNF Using the number o eligible women alling into speciic response categories, the eligible woman response rate (EWRR) is calculated as: 100 x EWC EWC + EWNH + EWR + EWPC + EWI + EWO 3 The overall response rate (ORR) is calculated as: Table A.3 Sample Implementation: men Percent distribution o households and eligible men in the DHS sample by result o the interview and household, eligible men and overall response rates, according to region and urban rural Area, Armenia 000 Result o interview and response rate Urban Rural Total Household interviews Completed (C) No competent respondent (HP) Reused (R) Dwelling not ound (DNF) Absent (A) Dwelling vacant (DV) Total percent Number 1,4 967,191 Response rate (HRR) Men interviews Completed (EMC) Not at home (EMNH) Reused (EMR) Partly completed (EMPC) Incapacitated (EMI) Total percent (EMO) Number 1, ,913 Response rate (EMRR) Overall response rate (ORR) Using the number o households alling into speciic response categories, the household response rate (HRR) is calculated as: 100 x C C + HP + R + DNF Using the number o eligible men alling into speciic response categories, the eligible man response rate (EMRR) is calculated as: 100 x EMC EMC + EMNH + EMR + EMPC + EMI + EMO 3 The overall response rate (ORR) is calculated as: ORR = HRR * EWRR/100 ORR = HRR * EMRR/100 * Appendix A