The Integration of the Israel Labour Force Survey with the National Insurance File

Transcription

1 The Integraton of the Israel Labour Force Survey wth the Natonal Insurance Fle Natale SHLOMO Central Bureau of Statstcs Kanfey Nesharm St. 66, corner of Bach Street, Jerusalem Abstact: In the year 2006, an Admnstratve Record Census wll be mplemented n Israel. The census, based on admnstratve fles, wll be combned wth large scale samplng and small area technques to enable accurate and legtmate counts. Gven the need to ncorporate admnstratve fles wth extensve survey samplng, we wll ntegrate the admnstratve data on the labour force n Israel from the Natonal Insurance Fle wth the Labour Force Survey, and thus calculate estmates for small areas whch otherwse could not be calculated due to the small sample szes. Keywords: ntegraton of admnstratve data, composte estmates, small area estmates 1. Introducton In the year 2006, an Admnstratve Record Census wll be mplemented n Israel. The census, based on admnstratve fles, wll be combned wth large scale samplng and small area technques to enable accurate and legtmate counts. The man admnstratve fles to be used for the census are the Natonal Populaton Regster, the Natonal Insurance Fle, the Property Tax Fle, and others. The fact that every ctzen n Israel s requred by law to have a natonal dentty number upon brth or upon mmgraton, greatly enhances the lnkage possbltes between the varous fles for the census. Gven the need to ncorporate admnstratve fles wth extensve survey samplng, we ntegrated the Natonal Insurance () Fle of 1995 from the Natonal Insurance Isttute wth the Labour Force Survey (LFS) Fle relevant to that same year. By ntegratng the fles from the year 1995, we are able to obtan updated addresses for the Fle by lnkng t to the last Populaton Census that was undertaken that same year, and to compare labour force characterstcs for small areas from the dfferent sources avalable. 441

2 Natale Shlomo Ths report wll gve an account of the results obtaned usng the geographcal areas of the cty of Jerusalem (excludng East Jerusalem), and compare estmates of labour force characterstcs for the regons and statstcal areas n the cty. 2. Preparaton of the Data 2.1 The Labour Force Survey The Labour Force Survey (LFS) carred out by Israel s Central Bureau of Statstcs s a cross-sectonal survey wth a rotatng panel desgn. The yearly sample of dwellngs s dvded nto four panels, whch are ntroduced gradually nto the feld, one panel every quarter, startng wth the last quarter of the samplng year. Every dwellng n each panel s nvestgated four tmes over a perod of a year and a half. The dwellng s nvestgated for two consecutve quarters, then rests for two quarters, and s nvestgated agan for an addtonal two quarters. Thus n every quarter there are four panels n the feld, each panel beng the frst, second, thrd or fourth ntervew of the dwellng. To supplement the sample of dwellngs, a separate sample was drawn from the nsttutonalzed populaton (boardng schools, student dormtores not n the major unverstes, dwellngs n sheltered-housng projects not ncluded n the Property Tax Fle, etc.) For the purpose of ths analyss we used the LFS Fles of the fourth quarter of 1995 and the frst quarter of The set of weghts for the households n the LFS Fle for each quarter was derved usng an teratve process based on the rakng method, n whch the dstrbuton of the weghted sample of the ndvduals s adjusted to ensure consstency wth external dstrbutons accordng to selected varables. The procedure assgns a sngle weght to the household, and controls for auxlary varables of the ndvduals. The auxlary varables used for the calculaton of weghts n the cty of Jerusalem were: One. Immgrant status Sex Age group (10 age groups) Two. Regon Sex Age group f the sample sze allowed t (3 age groups) The sample sze of the cty of Jerusalem (excludng East Jerusalem) for each quarter, was approxmately 1,600 ndvduals aged 15 and over n 720 households wth an average weght of about 165 (C.V. of the weghts, about 40%), and approxmately 380 ndvduals aged 15 and over sampled from nsttutons n the cty wth an average weght of about 45. The total number of persons aged 15 and over n the cty was about 290 thousand. The estmates provded here are an arthmetc average of the two quarters. Because of the rotatng panel method, half of the sample n the fourth quarter of 1995 were renvestgated n the frst quarter of The LFS Fle ncludes the addresses of the dwellngs n the sample but not ther statstcal area, so a specal lnkng procedure was used to obtan the statstcal areas for 442

3 The Integraton of the Israel Labour Force Survey wth the Natonal Insurance Fle each dwellng n the sample. Each address was parsed nto street names and house numbers. The street names were lnked to a fle that contans a key for transformng the street name nto a seral number, whch n turn was lnked to the Mnstry of Interor Electorate Votng Staton Fle, to obtan the statstcal area of each dwellng accordng to ther votng staton The Natonal Insurance Insttute Fle The Natonal Insurance Fle contans data on every employed person n Israel and ther wages, as reported to the Natonal Insurance Insttute. The fle also ncludes data on unemployment benefts and other stpends. Although the fle contans admnstratve data, t s subject to bases whch are typcal to fles of ths nature. One of the problems s that the addresses are not updated regularly. For ths analyss, the 1995 Fle was lnked to the 1995 Populaton Census by use of the dentty number. Ths was to obtan current up-dated addresses of the employed persons n Jerusalem, ncludng ther statstcal area. About seven percent of the persons n the Fle were not lnked wth the Census due to varous reasons, such as nonctzens who were employed n Israel, persons who were out-of scope at the tme of the census, and errors n the dentty number. Among the non-lnked employed persons n the Natonal Insurance Fle wth no known statstcal area, the statstcal area was mputed proportonal to the number of the employed persons wth known statstcal area The Census Sample Fle In 1995, Israel conducted the Populaton and Housng Census whch was a conventonal census wth full enumeraton. The Census also ncluded a large systematc sample of 20 percent of the populaton aged 15 and over, n whch the ndvduals were requested to fll out an extensve questonnare on ther socoeconomc characterstcs. About 15 percent of the total number of ndvduals n the sample underwent an mputaton procedure based on the hot deck method to compute for non-responses on labour force characterstcs. Snce the non-response s not random wthn the statstcal areas, we expect some bases on the data from the Census Sample (CS) Fle. 3. Results 3.1 Drect Estmates for the Annual Cvlan Labour Force Table 1 provdes the total number of employed persons (n hundreds) n the annual cvlan labour force by regons n Jerusalem accordng to the three sources: LFS, Fle and the CS Fle. As shown n Table 1, n some of the regons there are sgnfcant dscrepences between the Fle and the CS Fle, partcularly n the neghborhoods that are predomnantly 443

4 Natale Shlomo relgous where there was a large non-response rate for the Census, such as regons 2 and 6. Table 1: Total Number of Employed Persons n Annual Cvlan Labour Force (hundreds) Regon LFS CS Populat Sample Est. Rel. 2 s.e. 1 T Est. 2 s.e. on Sze error ot Sze (%) Total Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Composte Estmates for Regon Totals A composte estmate for the total number of employed persons n the annual cvlan labour force for regon, would be a weghted average of the estmate Ŷ LF from the LFS Fle and the total Ŷ from the Fle: LF Ŷ w Ŷ w 1 Ŷ where w s the weght for regon. al 444

5 The Integraton of the Israel Labour Force Survey wth the Natonal Insurance Fle There are several ways to determne the weghts, w, for a desgn-based model. One method for calculatng the weghts would be to base them on the estmates themselves. Although the estmates n the LFS have large varances, they have lttle bas because of the weghtng scheme, whereas the totals n the Fle have possbly large bases whch are hard to determne. Therefore, we consder the totals to be estmates. A possble weght for regon would be: LF s mn, Ŷ w 1. Ŷ Usng ths weght means that as long as the LFS estmate s greater than the estmate, we rely on the LFS estmate n the composte estmate, but when the LFS estmate s smaller, we borrow from the estmate. In the extreme case where the LFS estmate s zero, we rely solely on the estmate. Another weght, w, for regon, would be to take nto account the large varances of the sample survey estmates and the bases of the admnstratve data. Rao & Ghosh (1992) showed that an optmal weght, w o, can be obtaned by mnmsng the MSE of the composte estmate, Ŷ, wth respect to w o. Snce we beleve that there s no varance (or very lttle) n the admnstratve data, we obtan: o MSE Ŷ w LF. MSE Ŷ VAR Ŷ To estmate the bas of the total, Ŷ, we decded to use the CS estmates from the large sample of the Census. The CS estmates, despte the bases n the data, are more effcent because of ther small varances and therefore are more stable. An approxmately unbased estmator for the MSE of Ŷ s gven by : mse Ŷ CS 2 CS var Ŷ In four of the regons we obtaned an unstable estmate of the mse. In these cases, we pooled the regons wth ther neghborng regon and used a common weght for the pooled regon. Table 2 shows the results of the dfferent weghts used for the composte estmate and the relatve dfference n percentages (d) from the CS esmates. As expected, the optmum weghts gave smaller mean square error, and slghtly better results n comparson wth the CS estmates. Ŷ Ŷ. 445

6 Natale Shlomo Regon Table 2: Composte estmates for the Total Number of Employed Persons n the Annual Cvlan Labour Force (hundreds) Composte wth d Composte wth d w s doman sze w o optmum Est. 2 mse (%) Est. 2 mse (%) Total Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Regon Estmates for Statstcal Area Totals Snce the known populaton szes of the statstcal areas were not used as auxlary data n the weghtng scheme of the Labour Force Survey due to the small sample szes, we must frst compute the estmates of the statstcal areas. One way of dong ths s to produce a synthetc estmate out of the LFS estmate for the statstcal area j n regon, usng the known populaton szes, or: LF,S Nj LF Ŷ j N Ŷ where Nj s the known populaton sze n the statstcal area j belongng to regon. 446

7 The Integraton of the Israel Labour Force Survey wth the Natonal Insurance Fle Regon and Statstcal After ths adjustment, we can calculate the composte estmate for each statstcal area as shown n Secton 3.2, n whch we use the same optmum weght as regon for each separate statstcal area j, and replace the LFS estmate wth ts synthetc estmate. In Table 3, we show an example usng the sub-quarters 11 and 18, whch proved n the above tables to gve good results n comparson wth the CS estmates. Table 3: Small Area Estmates for the Total Number of Employed Persons n the Annual Cvlan Labour Force (hundreds) LFS Populaton synthetc Composte Sze LF I wth S S w o optmu d m Est. Est. (%) Area Es t. 2s.e. Est. Tota l Regon Area Area Area Area Area Regon Area Area Area Area Area Area Dscusson In ths analyss, we looked at a desgn based framework. Further analyss wll examne small area estmates for the statstcal areas n the regons usng model based desgns, as well as produce estmates for sub-domans, such as by sex. The lnkage technques used for ths analyss need to be perfected, snce despte the fact that every ctzen has an dentty number, we stll obtan a substantal number of non-lnks. Also, a much better understandng of the bases n the admnstratve data fles s necessary when preparng for the Admnstratve Record Census n 2006, especally the bases ncurred because of the mputaton methods for non-response. References: [1] (1) Ghosh, M. & Rao, J..N.K., Small Area Estmaton: An Apprasal, Statstcal Scence 1994, Vol. 9, No. 1, pp