Combining Landline Phone Samples and Mobile Phone Samples
Combining Landline Phone Samples and Mobile Phone Samples in Germany: a Dual Frame Approach First Outcomes of a DFG*-Project Siegfried Gabler, Sabine Häder *DFG = Deutsche Forschungsgemeinschaft = German Research Foundation
Overview • The main problem: The cell phone only households • Description of the DFG-Project • First results: - sampling procedure - weighting Gabler/Häder, Landline/cell phone, Q 2008 Rome 10 July 2008
Quelle: http: //www. bitkom. org/51940_50446. aspx Gabler/Häder, Landline/cell phone, Q 2008 Rome 10 July 2008
A challenge for survey methodology: The cell phone only households Slovakia 2008 55% Finland End of 2006 52% Poland 2007 24% France 2006 17% UK 2006 10% Germany 2007 8% Source: Lepkowski et al 2008; own investigation Gabler/Häder, Landline/cell phone, Q 2008 Rome 10 July 2008
DFG-Project: Interviews via cell phones • Cooperation TU Dresden and GESIS-ZUMA • Two parallel surveys: one via landline, one via cell phone • Aims of the project: - Investigation of mode effects - Response behaviour - Weighting model for the combination of both samples • Respectively ca. 1, 000 people interviewed Realization: Landline 1, 009 Cell phones 1, 161 • Survey period October 2007 -April 2008 Gabler/Häder, Landline/cell phone, Q 2008 Rome 10 July 2008
Sampling procedure for cell phone numbers by BIK n ASCHPURWIS + BEHRENS Gmb. H All hypothetical numbers assigned: 279, 000 numbers* Generated numbers on basis of used number blocks : 178, 050, 000 numbers Registered numbers 2, 297, 460 (January 2008) *) Provider O 2 allocates eight-digit numbers in connection with the prefix 0176. even though only seven-digit numbers were ordered at Reg. TP. Gabler/Häder, Landline/cell phone, Q 2008 Rome 10 July 2008
Dual frame survey Frame of landine phone numbers Frame of cell phone numbers household person Gabler/Häder, Landline/cell phone, Q 2008 Rome 10 July 2008
Availability: cell phone usage Gabler/Häder, Landline/cell phone, Q 2008 Rome 10 July 2008
i = 1. . . N (N is number of target persons in population Landline Cell phone MF quantity of numbers in frame MC quantity of numbers in frame m. F quantity of numbers in sample Selection: simple random sample m. C quantity of numbers in sample Selection: simple random sample ki. C Quantity of cell phone numbers allowing access to person i. ki. F Quantity of landline numbers allowing access to the household to which person i belongs. Selection: Kish table/last birthday zi Size of household to which person i belongs. Gabler/Häder, Landline/cell phone, Q 2008 Rome 10 July 2008
Weigthing model: basic assumption The probability of choosing two (not necessarily different) members of the same household via different modes is insignificant. (Could be a problem in small regional samples). Gabler/Häder, Landline/cell phone, Q 2008 Rome 10 July 2008
Inclusion Probability Horvitz-Thompson-Estimator Gabler/Häder, Landline/cell phone, Q 2008 Rome 10 July 2008
Weighting model Problem: Most of the household members do not know the exact quantity of landline numbers allowing access to their household. Therefore the simplified assumption: Landline (analog): 1 number Landline (ISDN): 2. 5 numbers ISDN = integrated services digital network ì 1 i. e. k = í î 2. 5 F i Gabler/Häder, Landline/cell phone, Q 2008 Rome 10 July 2008
Availability: Quantity of cell phone numbers Landline Cell phone n % One 771 88. 82 915 78. 74 Two 71 8. 18 192 16. 52 Three 19 2. 19 34 2. 93 Four 3 0. 35 9 0. 78 Five and more 4 0. 46 12 1. 03 ∑ 868 100. 0 1, 162 100. 0 missings 141 0 Gabler/Häder, Landline/cell phone, Q 2008 Rome 10 July 2008
Example DFG-Project Landline: MF = 129, 725, 800 m. F = 16, 154 Cell phone: MC = 178, 050, 000 m. C = 23, 955 n = 1, 009 + 1, 162 = 2, 171 Gabler/Häder, Landline/cell phone, Q 2008 Rome 10 July 2008
respondents 2, 171 landline phone onlys 141 analog 81 k. F = 1; k. C = 0 cell phone 2, 030 ISDN 59 k. F = 2. 5; k. C = 0 real landline phone 1, 778 analog 717 virtual landline phone 93 k. F = 2; k. C = v 4 no virtual landline phone 624 k. F = 1; k. C = v 4 no real landline phone 252 ISDN 1, 061 virtual landline phone 153 k. F = 3. 5; k. C = v 4 virtual landline phone 77 k. F = 1; k. C = v 4 cell phone onlys 175 k. F = 0; k. C = v 4 no virtual landline phone 908 k. F = 2. 5; k. C = v 4 = variable in data file Gabler/Häder, Landline/cell phone, Q 2008 Rome 10 July 2008
Histogram of weights 250 Cell phone Landline 200 Frequencies 150 100 50 0 0 2 4 6 weights Gabler/Häder, Landline/cell phone, Q 2008 Rome 10 July 2008 8
Following Steps • Evaluation • Publication of the results in Häder. M. und Häder. S. (Hrsg. ): Telefonbefragungen über das Mobilfunknetz. VS Verlag. Winter 2008. • Consecutive application for the DFG: calculation of the response rates • Conference in Mannheim about the results of the DFG-Project. November 14 th. 2008 • Further project within the framework of the DFGprogram „Survey Methodology“: Mobilpanel together with Marek Fuchs (University of Kassel) Gabler/Häder, Landline/cell phone, Q 2008 Rome 10 July 2008
- Slides: 17