1 2 Hybrid Model 4 JLPS Hybrid Model Hybrid Model Hybrid Model 1 2011; 1998, 2006; 1979; 2009 1999 1998, 2011 (Sociological Theory and Methods) 2012, Vol.27, No.1: 63-83 63
2011 2003; 2009 1 2 Allison2009 Hybrid Model Hybrid Model 2 2.1 Japanese Life course Panel Survey JLPS 2007 2011 5 4 25 JLPS 20 40 1) 2007 1 4,800 2) 36% 2 3,962 3 3,607 4 3,186 64
2.2 10 1 10 0 100 EGP Erikson et al. 1979EGP + +b++a4 EGP 4 0 3 5 7 9 SSM 1995 1998 36.7 10 3) 100 4) 3 2007 2010 1 1 1 1 2 1 2 N 1 N person-period 25 65
N 1 5021 47.099 18.413 0 100 5021 6.523 3.625 0 25 5021 1.588 0.934 0.140 5.340 5021 0.413 0.492 0 1 5021 0.140 0.347 0 1 5021 0.078 0.268 0 1 5021 0.369 0.483 0 1 2 1616 5.295 2.077 0 9 25-29 1616 0.254 0.436 0 1 30-34 1616 0.373 0.484 0 1 35-40 1616 0.373 0.484 0 1 1 4034 44.461 17.776 0 100 4034 6.601 4.086 0 25 4034 1.430 0.766 0.140 5.340 4034 0.294 0.456 0 1 4034 0.351 0.477 0 1 4034 0.030 0.170 0 1 4034 0.325 0.468 0 1 2 1364 4.939 1.640 0 9 25-29 1364 0.271 0.444 0 1 30-34 1364 0.322 0.467 0 1 35-40 1364 0.408 0.492 0 1 Source: JLPS2007-2010 2.3 Hybrid ModelAllison 2009 Hybrid Model Hybrid Model 5) [1] [1] Yit FE FEXit ui it FE (fixed effect model) i t X Y ui Y X it 1 ui Xit ui u i [1] FE ui Yit ( Y ) Yit [1] i i Y [2][1][2][3] [3] OLS 6) [2] Yi FE FEXi i 66
[3] Y Y ) ( X X ) ( ) ( it i FE it i it i ) [3] ( it i X Y [3] [3] X X ) Y Y ) ( it i ( it i X Y X Y FE [1]X X Y [1] [3] (a) (b) 1 2 1 2 (c) (d)hybrid Model 1 2 67
1 (a)2 5 (X)(Y) 1 X Y 2 X Y 1 (b)(c)(b) 2 2 1 (b) 1 (c) 5 X Y Hybrid Model Hybrid Model (random effect model) Hybrid Model Hybrid Model [4] Yit HM HM ( Xit Xi ) HM Xi HMZi ui it HM Hybrid Model Hybrid Model Y X ( Xit Xi ) X i X X X ) [4] Z i ( it i Hybrid Model Hybrid Model ui it 7) ui it 68
ui Hybrid Model [4] Hybrid Model Hybrid Model Hybrid Model X X HM X HM Hybrid Model X 1 (d)(a) Hybrid Model [4] HM [4] Y [5] Yi HM HMXi HMZi ui i HM X Y 1 Z 1 (d)2 5 5 Hybrid Model HM 1 (a)(d)(a) (d)(a) (c) FE Hybrid Model HM [4] Y Y ) [6] Y Y ) ( X X ) ( ) ( it i HM it i it i ( it i [3]FE HM 1 (d) 1 (d) [4] X Hybrid Model [7] 69
[7] Yit HM2 HM2Xit HM2Xi HM2Zi ui it 3 3.1 (variation) Null Model ICC 8) 0.53 0.56 4 5 0 100 Null Model 500 450 400 350 300 250 200 150 150 200 250 300 350 400 450 2 1 / 4 5 70
9) 1 2 3 a Hybrid Model Hybrid Model (with Centering) b (without Centering) b 1 0.5156 ** 0.1269 0.5156 ** 0.1270 0.5156 ** 0.1270 0.1153 0.7517 0.1153 0.7523 0.1153 0.7523 2.5174 + 1.3665 2.5174 + 1.3677 2.5174 + 1.3677-0.2763 1.4750-0.2763 1.4762-0.2763 1.4762-0.8868 1.9958-0.8868 1.9974-0.8868 1.9974 (omitted) (omitted) (omitted) Wave2-2.2267 ** 0.5046-2.2267 ** 0.5050-2.2267 ** 0.5050 Wave3 0.2201 0.5238 0.2201 0.5242 0.2201 0.5242 Wave4-0.0243 0.5548-0.0243 0.5553-0.0243 0.5553 2 0.7937 ** 0.1888 0.7937 ** 0.1888 1.2373 ** 0.1006 0.7217 ** 0.1620 1.2777 * 0.5047 1.1624 0.9059 9.3633 ** 1.1017 6.8459 ** 1.7562 4.8784 ** 1.1873 5.1547 ** 1.8944 4.2084 ** 1.4125 5.0952 * 2.4464 (omitted) (omitted) Wave2-5.0511 * 2.4878-2.8244 2.5386 Wave3-4.5596 + 2.4886-4.7798 + 2.5432 Wave4 3.7565 2.4321 3.7808 2.4947 25-29 -0.5209 0.8545-0.5209 0.8545 30-34 (omitted) (omitted) 35-40 1.0437 0.7599 1.0437 0.7599 47.0992 ** 0.1790 29.0884 ** 1.5187 29.0884 ** 1.5187 2( u ) 10.7039 0.2896 10.7039 0.2896 1( ) 12.6919 0.1534 12.6919 0.1534 0.4156 0.0155 0.4156 0.0155 2 19-20795.833 584.730-20795.833 584.730 Source: JLPS2007-2010 Note: ** : p < 0.01; * : p < 0.05; + : p < 0.1 a b Hybrid Model 3.2 71
2 1 13 1 2 2 10% 3 2 5% 10) 2003 2 1 1 1.24 1.87 11) 2 1 1 Hybrid Model 1 72
Hybrid Model 1 2 3.3 2 2 2 [4] 2 2 3 2 3 2 1 2 1 3 [7] 2 3 1 2 Snijders and Bosker 1999: pp.52-56 3 2 3 2 2 1 73
2 2000 10% 4 Hybrid Model Hybrid Model Hybrid Model 12) Hybrid Model Hybrid Model 2 Hybrid Model Hybrid Model 4.1 Hybrid Model 74
Hybrid Model 1000 1000 1000 X JLPS X JLPS Hybrid Model [8] Yit ( Xit Xi ) Xi Zi ui it Yit i t X Z ui it ui it ui N(0, 2 u)it N(0, 2 ) 2 2 u 2 2 ui it [8] 1000 75
Hybrid Model 95% 95% 1 0.5156 0.5144 0.0168 0.0162 0.939 0.5156 0.5144 0.0168 0.0161 0.938 0.1153 0.1013 0.5960 0.5663 0.943 0.1153 0.1013 0.5960 0.5649 0.942 2.5174 2.5267 2.0771 1.8717 0.936 2.5174 2.5267 2.0771 1.8671 0.935-0.2763-0.3300 2.2250 2.1805 0.946-0.2763-0.3300 2.2250 2.1751 0.946-0.8868-0.8939 3.9423 3.9923 0.949-0.8868-0.8939 3.9423 3.9825 0.947 Wave2-2.2267-2.2017 0.2599 0.2552 0.951-2.2267-2.2017 0.2599 0.2546 0.951 Wave3 0.2201 0.2378 0.2865 0.2750 0.946 0.2201 0.2378 0.2865 0.2743 0.945 Wave4-0.0243-0.0059 0.2976 0.3086 0.952-0.0243-0.0059 0.2976 0.3078 0.952 2 0.7937 0.7766 0.0348 0.0354 0.954 1.2373 1.2345 0.0095 0.0100 0.957 1.2777 1.2938 0.2579 0.2530 0.945 9.3633 9.3883 1.1934 1.2055 0.954 4.8784 4.8683 1.3692 1.4002 0.957 4.2084 4.2052 2.0301 1.9817 0.952 Wave2-5.0511-5.1129 5.9040 6.1559 0.958 Wave3-4.5596-4.5869 6.5873 6.1576 0.952 Wave4 3.7565 3.7571 6.0576 5.8797 0.950 25-29 -0.5209-0.5141 0.7813 0.7253 0.930 35-40 1.0437 1.0430 0.5921 0.5735 0.952 Source: JLPS2007-20101000 Note: 10002 210001000 10002 95%100095%1000 76
3 5 5 Hybrid Model Hybrid Model Hybrid Model 2 1000 E(ˆ ) 2 Hybrid Model 3 Hybrid Model 2 1 1000 Var(ˆ ) 1000 2 [9] E[ Var( ˆ) ] ^ 2 Hybrid Model Hybrid Model 1 10.1% 3 1000 95% 95% Pr[ L95% C. I. U 95% C. I. ] 95% t 95% 95% 3 95% 1 0.935 1 1.5% Hybrid Model 3 Hybrid Model 77
Hybrid Model Hybrid Model Hybrid Model Hybrid Model Hybrid Model Hybrid Model 9.9% 95% 0.936 1.4% Hybrid Model Hybrid Model 4.2 t z 0 [8] 0 [10] Y it 0 ( X X ) X Z u X Z u i it i i i it i i i X 1000 19 1000 19=19000 0 Var( ˆ 0) 3 it 78
[11] E[ Var( ˆ) ] ^ 0 0.5156 0 0 0 =0 =0 6.6%Wave2 0 5% t Hybrid Model 5% 0 1 5% 5%5% 5% 1 1 Wave2 0.066 5% 6.6% 1.6% Hybrid Model Hybrid Model =0 6.8%5% 1 5% 0.066 Hybrid Model 2 5 1 79
2 3 4 13) Hybrid Model Hybrid Model Hybrid Model Hybrid Model (i) (ii) (iii) Hybrid Model 2010Hybrid Model Hybrid Model Hybrid Model 4 Xi4 X 1 i Yi1 2011Hybrid Model 1810300322223005 1 GCOE 80
2 2 1) 2034 3540 2 2) 20082010 JLPS 1 36% 54% 3) 36.7 4) 4 1 5) Rabe-Hesketh and Skrondal2008 3 Allison2009 6) Stata xtreg fe [3] 7) Stata xtreg regls mle 8) Null Model 9) 2011 10) 1 3 F 2 11) 1 macro-effect 2 12) Hybrid Model Allison (2009: 27, note 9) balanced data balanced data OLS Hybrid Model GLS 81
13) 3 1 2 2 3 Allison, Paul. 2009. Fixed Effects Regression Models. Sage.. 2010.. Erikson, Robert, John H. Goldthorpe and Lucienne Portocarero. 1979. Intergenerational Class Mobility in Three Western European Societies: England, France and Sweden. British Journal of Sociology 30: 415-441.. 2003. : 105-126.. 2011. : 151-184.. 1998... 1999. 50(2): 216-230.. 2006... 2011. [3] : 63-77.. 2011. [3] : 95-110.. 2008. 2007.. 2009. JGSS 2000-2006 JGSS General Social Surveys [8]: JGSS JGSS : 1-12.. 1979. : 365-388. Rabe-Hesketh, Sophia and Anders Skrondal. 2008. Multilevel and Longitudinal Modeling Using Stata (second edition). Stata Press.. 2000. : 133-155. Snijders, Tom A. B. and Roel J. Bosker. 1999. Multilevel Analysis: An Introduction to Basic and Advanced Multilevel Modeling. Sage.. 2009... 1998. 1995 SSM [5] 1995 SSM. 82
. 2011... 2010. (JLPS).. 2011. JLPS. 2011 11 30 2012 3 6 The Within-Subject Effects of Intragenerational Class Mobility on Subjective Social Status Satoshi Miwa Koji Yamamoto Graduate School of Education Faculty of Policy Studies Tohoku University Chuo University 27-1 Kawauchi, Aoba-ku, Sendai-shi 742-1 Higashi-nakano, Hachioji-shi Miyagi, 980-8576, Japan Tokyo, 192-0393, Japan This article aimed to examine the determinants of subjective social status using panel/longitudinal data analyses. The first purpose was to illustrate the impact of class position on subjective social status in consideration of distinction between between-subject effects and within-subject effects. The second purpose was to show the analytical example of hybrid model, which means an application of multilevel modeling. Japanese Life course Panel Survey (JLPS) datasets drawn from first wave to fourth wave survey were used in empirical analyses. Fixed effect models and hybrid models were applied to those datasets. In addition, the performances of both models were compared by monte-carlo simulation method. The effect of intragenerational class mobility on subjective social status was partly supported in empirical analyses. Besides, some differences between within-subject regression coefficient and between-subject coefficient were found. According to the results of monte-carlo simulation, the performances of hybrid models were almost equivalent to those of fixed effect models. These findings gave suggestion that hybrid model is the one of useful options for analyzing panel/longitudinal data. Key words and phrases: Panel data, Multilevel analysis, Fixed effect model, Sujective social status (Received November 30, 2011/ Accepted March 6, 2012) 83
84