on the benefits of using the process macro to · pdf fileon the benefits of using the process...

Download ON THE BENEFITS OF USING THE PROCESS MACRO TO · PDF fileon the benefits of using the process macro to test statistical models ... ***** index of moderated mediation ... (especially

If you can't read please download the document

Upload: trandang

Post on 06-Feb-2018

224 views

Category:

Documents


3 download

TRANSCRIPT

  • ON THE BENEFITS OF USING THE PROCESS MACRO TO TEST STATISTICAL MODELS IN DATA

    ANALYSIS CLASSES

    Renaud Lunardo, PhD, HDR

    Kedge Business School, Bordeaux

    [email protected]

    Xlstat Conference, Bordeaux, Juin 2017

  • Xlstat Conference, Bordeaux, Juin 2017

    What students learn in statistical classes

    X Y

    Basic tests :- T-tests, ANOVAs- Correlations, regressions

    1/ Descriptive statistics: - means, medians, quartiles, proportions, confidence intervals

    2/ Bivariate / Multivariate statistics:

  • Xlstat Conference, Bordeaux, Juin 2017

    Beyond main effects: moderation and mediation

    X

    Z

    Y X

    M

    Y

    GOAL: identify the conditions under which:- X has an effect on Y- or the effect of X on Y changes

    GOAL: identify the mechanism that explainswhy X has an effect on Y

  • Xlstat Conference, Bordeaux, Juin 2017

    A step further: Moderated medations

    X

    M

    Y

    Z

  • Xlstat Conference, Bordeaux, Juin 2017

    Testing main effects using SEMs: General principles and vocabulary

    Latent variable

    Indicator 1E 1

    Indicator 2

    Indicator 3

    Indicator 4

    E 2

    E 3

    E 4

    1

    2

    3

    4

    1

  • Xlstat Conference, Bordeaux, Juin 2017

    SEMs enable the inclusion of second-ordervariables (multidimensional scales)

    Constructive

    X11

    X2

    X3

    X4

    11

    21

    31

    41

    12

    3

    4

    Offensive

    X55

    X6

    X7

    X8

    52

    62

    72

    82

    26

    7

    8

    Humor

  • Xlstat Conference, Bordeaux, Juin 2017 7

    Notations : Independent variable Dependent variable

    Indicator X Y

    Latent variable (ksi) (ta)

    Error in the measure of

    indicators (delta) (epsilon)

    Loadings (PCA) (lambda)

    Relation (gamma) (bta)

    Covariance (phi) (psi)

    Error of specification of

    the model- (zta)

    General principles of SEMs

  • Xlstat Conference, Bordeaux, Juin 2017

    Testing models using SEMS

    11

    21

    12

    22

    12

    Performance

    Y1 1

    Y2

    Y3

    Y4

    11

    21

    31

    41

    12

    3

    4

    1

    Trust

    Y5 5

    Y6

    Y7

    Y8

    52

    62

    72

    82

    26

    7

    82

    Humor

    X11

    X2

    X3

    X4

    11

    21

    31

    41

    12

    3

    4

    Competence

    X55

    X6

    X7

    X8

    52

    62

    72

    82

    26

    7

    8

  • Xlstat Conference, Bordeaux, Juin 2017

    Testing moderations using SEMs: the multigroups analyses

    11

    21

    12

    22

    12

    Benevolence

    Integrity

    Purchase

    Attitude

    Past experience with the brand

  • Xlstat Conference, Bordeaux, Juin 2017

    the (numerous) assumptions behind them

    Measurement invariance Goal: is the latent factor structure underlying the items measures the same across groups (Judd and Kenny 2010).Configural invariance: same structure across groups

    Error variance invariance: same error variance across groups

    Metric invariance: same factor loadings across groups

    Factor variance invariance: same factor variance across groups

    Scalar invariance: same item intercepts across groups

    Factor covariance invariance: same factor covariance across groups

    Factor mean invariance: same factor mean across groups

  • Xlstat Conference, Bordeaux, Juin 2017

    and its drawbacks

    1. Boostrap

    2. Often need to perform a split (median- or mean-based)

    Huge controversy around this method due to, among others, arbitrary value of the split and insensitive analysis to the pattern of local covariation between X and Y within groups defined by the median split

    We know of no statistical argument in favor of median splits to counterbalance the chorus of statistical critiques against them (Mc Lelland et al., 2015, p.680).

  • Xlstat Conference, Bordeaux, Juin 2017

    The need to check the fit of the model

    Fit Formula

    Bentler-Bonett (NFI)2(Null Model) - 2(Proposed Model)

    ----------------------------------------------------2(Null Model)

    Tucker-Lewis Index (TLI)2/df(Null Model) - 2/df(Proposed Model)

    ---------------------------------------------------------------2/df(Null Model) - 1

    CFId(Null Model) - d(Proposed Model)

    -----------------------------------------------------d(Null Model)

    RMSEA(2 - df)

    ---------------------[df(N - 1)]

  • Xlstat Conference, Bordeaux, Juin 2017

    And the light comes: the Process macro

  • Xlstat Conference, Bordeaux, Juin 2017

    How to use it: easy

  • Xlstat Conference, Bordeaux, Juin 2017

    An easy-to-interpret output

    Outcome: CONF

    coeff se t p LLCI ULCIconstant 2,6147 ,2505 10,4367 ,0000 2,1206 3,1088EXPLOR 1,5231 ,7383 2,0630 ,0404 ,0670 2,9793HUMOR ,3109 ,0653 4,7635 ,0000 ,1822 ,4396int_1 -,4290 ,1910 -2,2460 ,0258 -,8057 -,0523

    R-square increase due to interaction(s):R2-chng F df1 df2 p

    int_1 ,0231 5,0443 1,0000 195,0000 ,0258

    *************************************************************************

    Conditional effect of X on Y at values of the moderator(s):EXPLOR Effect se t p LLCI ULCI,0000 ,3109 ,0653 4,7635 ,0000 ,1822 ,43961,0000 -,1181 ,1795 -,6581 ,5113 -,4722 ,2359

    1

    1,5

    2

    2,5

    3

    3,5

    4

    4,5

    1 2 3 4 5 6 7

    Tru

    st

    Use of humor

    Not exploration phase Exploration phase

  • Xlstat Conference, Bordeaux, Juin 2017

    Moderations using Process: Floodlight analyses

  • Xlstat Conference, Bordeaux, Juin 2017

    A concrete example of floodlight analysis

    1

    2

    3

    4

    5

    6

    7

    1 2 3 4 5 6 7

    Po

    siti

    ve R

    ein

    terp

    reta

    tio

    n

    Rumination

    Low Guilt

    High Guilt

    Outcome: REINTERP

    coeff se t p

    constant 2,1844 ,3294 6,6324 ,0000

    RUMIN ,5474 ,1158 4,7283 ,0000

    GUILTMAN 2,9101 ,5617 5,1805 ,0000

    int_1 -,5843 ,1736 -3,3651 ,0009

    ******************************************************

    Effect of X on Y at values of the moderator(s):

    RUMIN Effect se t p

    1,4270 2,0764 ,3551 5,8471 ,0000

    2,8340 1,2542 ,2439 5,1414 ,0000

    4,2411 ,4320 ,3351 1,2893 ,1990

    ********** JOHNSON-NEYMAN TECHNIQUE *****************

    Moderator value(s) defining Johnson-Neyman significance

    region(s):

    Value % below % above

    3,9561 78,5311 21,4689

  • Xlstat Conference, Bordeaux, Juin 2017

    More in details

    Conditional effect of X on Y at values of the moderator (M)

    RUMIN Effect se t p LLCI ULCI

    1,0000 2,3258 ,4122 5,6424 ,0000 1,5122 3,1394

    1,3000 2,1505 ,3715 5,7895 ,0000 1,4174 2,8837

    1,6000 1,9753 ,3339 5,9161 ,0000 1,3163 2,6342

    1,9000 1,8000 ,3007 5,9869 ,0000 1,2065 2,3934

    2,2000 1,6247 ,2734 5,9429 ,0000 1,0851 2,1643

    2,5000 1,4494 ,2540 5,7065 ,0000 ,9481 1,9507

    2,8000 1,2741 ,2444 5,2141 ,0000 ,7918 1,7564

    3,1000 1,0988 ,2456 4,4734 ,0000 ,6140 1,5836

    3,4000 ,9235 ,2577 3,5843 ,0004 ,4150 1,4321

    3,7000 ,7482 ,2790 2,6813 ,0080 ,1974 1,2990

    3,9561 ,5986 ,3033 1,9738 ,0500 ,0000 1,1971

    4,0000 ,5729 ,3079 1,8610 ,0644 -,0347 1,1806

    4,3000 ,3976 ,3422 1,1620 ,2468 -,2778 1,0731

  • Xlstat Conference, Bordeaux, Juin 2017

    Turning to mediations : Focus on CIsOutcome: MPERF

    coeff se t p

    constant 2,9574 ,2648 11,1676 ,0000

    MCONF ,2241 ,0741 3,0259 ,0029

    *******************************************************

    Outcome: MBAO

    Coeff se t p

    constant 1,6966 ,2740 6,1919 ,0000

    MPERF ,2674 ,0598 4,4738 ,0000

    MCONF ,3590 ,0601 5,9764 ,0000

    ******************** DIRECT AND INDIRECT EFFECTS ******

    Direct effect of X on Y

    Effect SE t p

    ,3590 ,0601 5,9764 ,0000

    Indirect effect of X on Y

    Effect Boot SE BootLLCI BootULCI

    MPERF ,0599 ,0285 ,0181 ,1327

  • Xlstat Conference, Bordeaux, Juin 2017

    Mediations using Process: Even with 2 mediators

    *************** DIRECT AND INDIRECT EFFECTS ******************

    Direct effect of X on Y

    Effect SE t p LLCI ULCI

    -,0588 ,2515 -,2340 ,8154 -,5567 ,4390

    Indirect effect(s) of X on Y

    Effect Boot SE BootLLCI BootULCI

    Total: -,0956 ,1362 -,4039 ,1294

    Ind1 : ,1383