% % 34

* 2000 2005 1% 1% 1% 1% * VZDA2010-15 33

2011. 3 2009 2009 2004 2008 1982 1990 2000 2005 1% 1 1 2005 1% 34

2000 2005 1% 35

2011. 3 2000 0. 95 20-30 209592 70982 33. 9% 2005 1% 258 20-30 372301 115483 31. 0% 2000 2005 0. 79% 1. 70% 2000 33. 49% 33. 33% 14. 93% 11. 98% 2005 2000 2005 2005 20 51. 63% 25 32. 32% 30 18. 73% 36

2000 2005 20-30 2005 1% 51 26. 3 51. 64% 15. 7 30. 84% 2005 1% 20-30 37

2011. 3 2005 1% 764195 65. 2% 34. 8% 12. 43% 5. 47% 5. 99% 4. 31% 38

/ / 1 1 2010 2005 1% logit regression 39

2011. 3 2005 1% 1975-1985 118121 95075 2 2 3889 3. 29 / 252. 21 104463 88. 44 6691 5. 66 1 1720 1. 46 262. 22 844. 71 118121 100. 00 3 7 19615 1 40

3 % 1975-1979 1980-1985 1975-1979 1980-1985 51. 5 50. 2 51. 8 65. 04 78. 16 60. 03 10. 1 10. 4 10 12. 56 12. 51 12. 58 1. 6 2. 5 1. 4 2. 12 2. 58 1. 94 12. 6 17. 2 11. 6 13. 41 16. 75 12. 13 47. 7 46 48. 1 54. 65 57. 09 53. 71 18. 9 17. 6 19. 2 17. 70 14. 74 18. 83 11. 4 10. 1 11. 8 7. 70 5. 81 8. 42 7. 3 5. 9 7. 7 4. 21 2. 74 4. 76. 3. 8. 3. 23. 29. 20 3. 3 3. 1 3. 3 2. 68 2. 65 2. 69 6. 7 7 6. 7 6. 06 6. 20 6. 01 7 7. 8 6. 8 4. 79 4. 75 4. 81 10. 9 9. 8 11. 1 9. 50 7. 90 10. 12 51. 1 55. 3 50. 2 61. 78 67. 12 59. 73 21 17. 1 21. 9 15. 19 11. 38 16. 65 32. 2 31. 7 32. 4 21. 99 19. 52 22. 93 2000 4 4. 3 3. 9 5. 03 4. 20 5. 35 7. 29 6. 93 7. 38 7. 92 7. 25 8. 18 2. 80 3. 00 2. 74 3. 24 3. 33 3. 17 N 19615 95075 95075 7 1975-1980 78. 16% 10. 1% 12. 56% 19% 12. 14% 41

2011. 3 10 4 1 2005 1% 31% 1 3 42

43

2011. 3 44

118121 259545 14. 5% 14. 13% 5 66. 61% 27. 57% 5. 82% 20-30 5 % 66. 61 2. 68 69. 29 3. 14 27. 57 30. 71 69. 75 30. 25 100. 00 6 15. 68% 45

2011. 3 6 116488 44. 88 9854 3. 80 40703 15. 68 46098 17. 76 8221 3. 17 32554 12. 54 5627 2. 17 259545 100. 00 1 - a 1 - b 1 - c 1 - d 46

1 6 6 6 7 8 47

2011. 3 48

49

2011. 3 8 4 5 B Exp B S. E. B Exp B S. E. 1. 742 ** 5. 711. 101 1. 745 ** 5. 724. 064 1. 406 ** 4. 079. 096 1. 148 ** 3. 152. 055 1. 752 ** 5. 767. 095 1. 628 ** 5. 093. 056 1. 431 ** 4. 184. 095 1. 086 ** 2. 963. 052 1. 220 ** 3. 389. 093. 856 ** 2. 355. 050. 252 ** 1. 286. 007. 247 ** 1. 280. 006. 762 ** 2. 143. 047. 668 ** 1. 951. 047 1. 512 ** 4. 535. 067 1. 228 ** 3. 415. 038. 390 ** 1. 476. 039. 198 ** 1. 219. 029 -. 286 **. 751. 088 -. 253 **. 776. 060 1980-1985. 930 ** 2. 533. 113. 830 ** 2. 293. 072 * -. 332 **. 717. 122 -. 303 **. 739. 082 * -. 164. 849. 115 -. 268 **. 765. 070 * -. 067. 935. 114 -. 374 **. 688. 071 * -. 266 *. 767. 112 -. 076. 927. 063 *. 110 1. 116. 108. 143 * 1. 154. 059 * -. 045 **. 956. 009 -. 055 **. 947. 007 *. 014 1. 014. 056 -. 118 *. 888. 058 *. 536 ** 1. 709. 078. 375 ** 1. 454. 046 * -. 315 **. 730. 046 -. 291 **. 747. 036 * -. 266 *. 767. 106 -. 154 *. 857. 074-8. 008 **. 000. 093-6. 183 **. 002. 057-2 log likelihood 66048. 086 101000. 534 N 95075 91263 * p < 0. 05 **p < 0. 01. 3 3 50

9 6 6 51

2011. 3 2000 2005 1% 20-30 2 /3 1 /3 1% 52

2005 1% 2000 2005 1% 2010 3 2009 80 3 2009 1990-2000 5 2004 5 2008 5 53

SOCIOLOGICAL STUDIES Bimonthly Vol. 26 2011 3 May 2011 PAPER Causal Analyses in Social Sciences Peng Yusheng 1 Abstract Hume s problem of causal induction has inspired more than two centuries of epistemological discussion. J. S. Mill was the first one to elaborate the methodological principles of causal induction. The Millian principle of overall similarities for causal inference laid the foundation for Fisher s randomized experimental design and guided causal analyses in multivariate statistical modeling and qualitative comparative case studies. Causal theories are not however induced from empirical correlations. They are produced through leaps of faith and empirically tested through deductive logic. The best research strategy is triangulation that combines theorizing qualitative analysis and statistical modeling. On the Misusage and Adjustment of Household Members Matched Data A discussion with the paper Expansion of Higher Education and Inequality in Opportunity of Education Yang Ge & Wang Guangzhou 33 Abstract The data of household members matched are widely used in sociology demography and related research but the selection bias of the matched data is often ignored. This paper uses the raw data of the Fifth Census in 2000 and 1% Population Sample Survey in 2005 to match three kinds of relations i. e. father and sons mother and children husband and wife and confirms that the selection bias exist in these matched data grouped by age gender migration status rural and urban distribution education geographical distribution and so on. Based on the matched data this paper re-tests the matched data analysis models and conclusions of the paper Expansion of Higher Education and Inequality in Opportunity of Education and finds out that the selection bias of matched data would influence the accuracy of model a- nalysis and research conclusion. For further reducing the impact of matched data bias we propose two adjustment methods and find out that re-sampling and weighting methods can reduce the selective bias. Selection Bias and Treatment Methods A response to the questions 243