Nested Example Using SPSS David A. Kenny January 8, 2014.

29
Nested Example Using SPSS David A. Kenny January 8, 2014

Transcript of Nested Example Using SPSS David A. Kenny January 8, 2014.

Page 1: Nested Example Using SPSS David A. Kenny January 8, 2014.

Nested Example Using SPSS

David A. Kenny

January 8, 2014

Page 2: Nested Example Using SPSS David A. Kenny January 8, 2014.

Presumed Background

• Multilevel Modeling

Page 3: Nested Example Using SPSS David A. Kenny January 8, 2014.

Example Kashy (1991) Study of Gender and Intimacy

respondents completed a survey each night for two weeks

outcome is the average intimacy rating of each interaction partner(from 1 to 7, bigger numbers more intimacy)

Levels level 1: intimacy of the interaction (1-7),

partner gender (-1=male; 1=female) level 2: respondent gender (-1=male;

1=female)   3

Page 4: Nested Example Using SPSS David A. Kenny January 8, 2014.

DownloadDataSyntaxOutput

4

Page 5: Nested Example Using SPSS David A. Kenny January 8, 2014.

5

Page 6: Nested Example Using SPSS David A. Kenny January 8, 2014.

Equations A “separate” regression equation for each level 2 unit:Level 1 equationIntimacy = b0(intercept) + b1(partner gender) + error1

The coefficients from the level 1 equation become the “dependent” variables:Level 2 equations•b0 = “average” intercept + effect of respondent gender + error2

•b1 = “average” effect of partner gender + effect of respondent gender + error3

Note that the effect respondent gender on the slope, b1, for partner gender is the interaction of the two gender variables. 

6

Page 7: Nested Example Using SPSS David A. Kenny January 8, 2014.

Predicting Intimacy with Partner’s Gender for Each Participant

Men

ID Intercept (b0i) Slope (b1i) Number of Partners

1 5.35 .76 11

2 3.39 -.14 8

….

26 4.41 .37 14

Mean 3.85 .24

Women

27 4.49 -.11 35

28 4.03 .03 22

77 4.40 .32 19

Mean 4.39 -.16

7

Page 8: Nested Example Using SPSS David A. Kenny January 8, 2014.

EffectsFixed Effects

“average” intercept (b0; like a grand mean) effect of respondent gender “average” slope (b1; partner gender) interaction of partner and respondent

genderRandom effects

variance error variance intercept or b0 variance slope or b1 variance

covariance: intercept with slope8

Page 9: Nested Example Using SPSS David A. Kenny January 8, 2014.

Centering and the Example: Effects Coding

Partner gender and respondent gender effects coded (-1 = male, +1 = female):

• overall intercept: respondents’ typical level of intimacy across both females and males

• intercept variance: differences in respondent’s typical level of intimacy across females and males

• overall slope: overall effect of partner gender across female and male respondents

• slope variance: differences in the effect of partner gender

Note with effects coding, all effects are one-half the relative “advantage” or “disadvantage” of females over males because the difference between females and males is two units.

9

Page 10: Nested Example Using SPSS David A. Kenny January 8, 2014.

10

Page 11: Nested Example Using SPSS David A. Kenny January 8, 2014.

11

Page 12: Nested Example Using SPSS David A. Kenny January 8, 2014.

12

Page 13: Nested Example Using SPSS David A. Kenny January 8, 2014.

13

Page 14: Nested Example Using SPSS David A. Kenny January 8, 2014.

14

Page 15: Nested Example Using SPSS David A. Kenny January 8, 2014.

15

Page 16: Nested Example Using SPSS David A. Kenny January 8, 2014.

16

Page 17: Nested Example Using SPSS David A. Kenny January 8, 2014.

Syntax

MIXED

intimacy WITH resp_gender partner_gender

/FIXED = resp_gender partner_gender resp_gender*partner_gender

/PRINT = SOLUTION TESTCOV

/RANDOM INTERCEPT partner_gender | SUBJECT(id) COVTYPE(UNR).

17

Page 18: Nested Example Using SPSS David A. Kenny January 8, 2014.

Random Effects

18

Page 19: Nested Example Using SPSS David A. Kenny January 8, 2014.

Testing Variances in SPSS

• In SPSS all tests of variances are two-tailed.

• There is no interest in whether the variance is less than zero (in fact, the variance cannot never be less than zero).

• We can cut the p value in half for the variances.

19

Page 20: Nested Example Using SPSS David A. Kenny January 8, 2014.

Example: Random Effects (in words)

There is variation in the intercept: Some people say that they are more intimate than do others. Proportion of variance (intraclass correlation) due to the intercept: .852973/(.852973+ 1.890825) = .311.

Variation due to partner gender not significant (p = .167) and could be dropped from the model.

20

Page 21: Nested Example Using SPSS David A. Kenny January 8, 2014.

Example: Fixed Effects

21

df can be non-integer!

Page 22: Nested Example Using SPSS David A. Kenny January 8, 2014.

Fixed Effects (in words)

Females say that their interactions are more intimate than males by about half a point. (Remember with effects coding the difference between a man and a woman is two.)

People say interactions with females are more intimate by about a tenth of a point, but this difference is not statistically significant.

Mixed-gendered interactions (MF & FM) are viewed as more intimate than same-gendered interactions (MM & FF) by about a third of a point.

22

Page 23: Nested Example Using SPSS David A. Kenny January 8, 2014.

Cell Means

/EMMEANS=TABLES(overall) WITH (resp_gender=1 partner_gender=1)

/EMMEANS=TABLES(overall) WITH (resp_gender=1 partner_gender=-1)

/EMMEANS=TABLES(overall) WITH (resp_gender=-1 partner_gender=1)

/EMMEANS=TABLES(overall) WITH (resp_gender=-1 partner_gender=-1)

23

Page 24: Nested Example Using SPSS David A. Kenny January 8, 2014.

24

Page 25: Nested Example Using SPSS David A. Kenny January 8, 2014.

25

Page 26: Nested Example Using SPSS David A. Kenny January 8, 2014.

Fractional Degrees of Freedom

Degrees of freedom are fractional because standard errors are variances that are pooled across levels.

Method called Satterthwaite approximation.

26

Page 27: Nested Example Using SPSS David A. Kenny January 8, 2014.

Centering and the Example: Dummy Coding

Partner gender and respondent gender dummy coded (0: males; +1: females):

• overall intercept: male respondents’ typical level of intimacy with male partners

• intercept variance: differences in respondent’s typical level of intimacy with male partners

• overall slope: effect of partner gender for male respondents

Note with dummy coding, all effects are the relative “advantage” or “disadvantage” of female over males.

27

Page 28: Nested Example Using SPSS David A. Kenny January 8, 2014.

Thanks! Debby Kashy

28

Page 29: Nested Example Using SPSS David A. Kenny January 8, 2014.

29

More WebinarsReferences

Programs

Growth Curve

Repeated Measures

Two-Intercept Model

Crossed Design

Other Topics