------------------------------------------------------------------------------------------------------------------------------------------------------
name: STATA_Chapter12
log: C:\Dropbox\PilesOfVariance\Chapter12\STATA\STATA_Chapter12_Output.smcl
log type: smcl
opened on: 21 Oct 2014, 10:32:04
.
. * Import chapter 12 stacked data
. use "$filesave\STATA_Chapter12.dta", clear
.
. * Collapse data into one row per person to summarize subject variables
. display as result "Chapter 12: Descriptive Statistics for Subject-Level Variables"
Chapter 12: Descriptive Statistics for Subject-Level Variables
. collapse older age, by(subjectid)
. tabulate older, summarize(age)
(mean) | Summary of (mean) age
older | Mean Std. Dev. Freq.
------------+------------------------------------
0 | 19.708333 2.2660732 96
1 | 75.701754 5.3684426 57
------------+------------------------------------
Total | 40.568627 27.414164 153
.
. * Re-import chapter 12 stacked data
. use "$filesave\STATA_Chapter12.dta", clear
.
. * Collapse data into one row per item to summarize item variables
. display as result "Chapter 12: Descriptive Statistics for Item-Level Variables"
Chapter 12: Descriptive Statistics for Item-Level Variables
. collapse relevance salience, by(itemid)
. summarize(relevance salience)
Variable | Obs Mean Std. Dev. Min Max
-------------+--------------------------------------------------------
relevance | 51 2.647059 1.906028 0 5
salience | 51 3.019608 1.104359 1 5
.
. * Re-import chapter 12 stacked data and create variables for analysis
. use "$filesave\STATA_Chapter12.dta", clear
. * Log of response time
. gen logrt = log(rt)
(157 missing values generated)
. * Make copy of 'older' and dummy codes to be used in heterogeneous variance models
. gen agegroup = older
. gen isyounger=0
. replace isyounger=1 if (older==0)
(4896 real changes made)
. gen isolder=1
. replace isolder=1 if (older==1)
(0 real changes made)
. * Piecewise effect of age
. gen yrs65=0
. replace yrs65=age-65 if (older==1)
(2856 real changes made)
. * Item predictors
. gen rel3 = relevance - 3
. gen sal3 = salience - 3
. * Item salience random slope for heterogeneous random effects by age
. gen isysal = isyounger*sal3
. gen isosal = isolder*sal3
. label variable logrt "logrt: Natural Log RT in Seconds"
. label variable agegroup "agegroup: Younger=0, Older=1"
. label variable isyounger "isyounger: 0=Older, 1=Younger"
. label variable isolder "isolder: 0=Younger, 1=Older"
. label variable yrs65 "yrs65: Subject Years of Age Over 65 (0=65)"
. label variable rel3 "rel3: Item Relevance (0=3)"
. label variable sal3 "sal3: Item Salience (0=3)"
. label variable isysal "isysal: Younger Random Salience Slope"
. label variable isosal "isosal: Older Random Salience Slope"
.
. * Subset sample to complete cases for all predictors
. egen nummiss = rowmiss(older yrs65 relevance salience logrt)
. drop if nummiss>0
(157 observations deleted)
.
. display as result "Chapter 12: Descriptive Statistics for Trial-Level Variables"
Chapter 12: Descriptive Statistics for Trial-Level Variables
. summarize(rt logrt)
Variable | Obs Mean Std. Dev. Min Max
-------------+--------------------------------------------------------
rt | 7646 7.376688 8.08914 1.156 59.969
logrt | 7646 1.612058 .8284638 .1449658 4.093828
.
. ********************************************************************************
. ******* BEGIN CHAPTER 12 CROSSED RANDOM EFFECTS MODELS *******
. ******* NOTE: MODELS WITH RANDOM SLOPES WILL NOT CONVERGE *******
. ******* BUT ARE PROVIDED BELOW FOR ILLUSTRATION *******
. ********************************************************************************
.
. display as result "Eq 12.8: Empty Means, Single-Level Model for Log RT"
Eq 12.8: Empty Means, Single-Level Model for Log RT
. mixed logrt , ///
> || subjectid: , noconstant variance reml covariance(unstructured),
Note: all random-effects equations are empty; model is linear regression
Mixed-effects REML regression Number of obs = 7646
Wald chi2(0) = .
Log restricted-likelihood = -9413.6033 Prob > chi2 = .
------------------------------------------------------------------------------
logrt | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
_cons | 1.612058 .0094745 170.15 0.000 1.593488 1.630627
------------------------------------------------------------------------------
------------------------------------------------------------------------------
Random-effects Parameters | Estimate Std. Err. [95% Conf. Interval]
-----------------------------+------------------------------------------------
var(Residual) | .6863522 .0111006 .6649367 .7084574
------------------------------------------------------------------------------
. estat ic, n(1),
Akaike's information criterion and Bayesian information criterion
-----------------------------------------------------------------------------
Model | Obs ll(null) ll(model) df AIC BIC
-------------+---------------------------------------------------------------
. | 1 . -9413.603 2 18831.21 18827.21
-----------------------------------------------------------------------------
Note: N=1 used in calculating BIC
. estimates store FitEmpty1,
.
. display as result "Eq 12.9: Add Subject Random Intercept Variance"
Eq 12.9: Add Subject Random Intercept Variance
. mixed logrt , ///
> || subjectid: , variance reml covariance(unstructured),
Note: single-variable random-effects specification in subjectid equation; covariance structure set to identity
Performing EM optimization:
Performing gradient-based optimization:
Iteration 0: log restricted-likelihood = -8540.3628
Iteration 1: log restricted-likelihood = -8540.3628
Computing standard errors:
Mixed-effects REML regression Number of obs = 7646
Group variable: subjectid Number of groups = 153
Obs per group: min = 43
avg = 50.0
max = 51
Wald chi2(0) = .
Log restricted-likelihood = -8540.3628 Prob > chi2 = .
------------------------------------------------------------------------------
logrt | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
_cons | 1.619391 .0347712 46.57 0.000 1.55124 1.687541
------------------------------------------------------------------------------
------------------------------------------------------------------------------
Random-effects Parameters | Estimate Std. Err. [95% Conf. Interval]
-----------------------------+------------------------------------------------
subjectid: Identity |
var(_cons) | .1746526 .0212411 .1376108 .2216652
-----------------------------+------------------------------------------------
var(Residual) | .5157229 .0084258 .4994702 .5325045
------------------------------------------------------------------------------
LR test vs. linear regression: chibar2(01) = 1746.48 Prob >= chibar2 = 0.0000
. estat ic, n(1),
Akaike's information criterion and Bayesian information criterion
-----------------------------------------------------------------------------
Model | Obs ll(null) ll(model) df AIC BIC
-------------+---------------------------------------------------------------
. | 1 . -8540.363 3 17086.73 17080.73
-----------------------------------------------------------------------------
Note: N=1 used in calculating BIC
. estat icc,
Intraclass correlation
------------------------------------------------------------------------------
Level | ICC Std. Err. [95% Conf. Interval]
-----------------------------+------------------------------------------------
subjectid | .252982 .0232155 .2102295 .3011139
------------------------------------------------------------------------------
. estimates store FitEmpty2,
.
. display as result "Eq 12.10: Add Item Random Intercept Variance"
Eq 12.10: Add Item Random Intercept Variance
. mixed logrt , ///
> || _all: R.subjectid , variance reml ///
> || _all: R.itemid ,
Performing EM optimization:
Performing gradient-based optimization:
Iteration 0: log restricted-likelihood = -7591.0186
Iteration 1: log restricted-likelihood = -7591.0186
Computing standard errors:
Mixed-effects REML regression Number of obs = 7646
Group variable: _all Number of groups = 1
Obs per group: min = 7646
avg = 7646.0
max = 7646
Wald chi2(0) = .
Log restricted-likelihood = -7591.0186 Prob > chi2 = .
------------------------------------------------------------------------------
logrt | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
_cons | 1.622606 .0608129 26.68 0.000 1.503414 1.741797
------------------------------------------------------------------------------
------------------------------------------------------------------------------
Random-effects Parameters | Estimate Std. Err. [95% Conf. Interval]
-----------------------------+------------------------------------------------
_all: Identity |
var(R.subjec~d) | .1802584 .0215904 .1425421 .2279543
-----------------------------+------------------------------------------------
_all: Identity |
var(R.itemid) | .1259187 .0257051 .0843968 .1878688
-----------------------------+------------------------------------------------
var(Residual) | .3898709 .006391 .3775438 .4026006
------------------------------------------------------------------------------
LR test vs. linear regression: chi2(2) = 3645.17 Prob > chi2 = 0.0000
Note: LR test is conservative and provided only for reference.
. estat ic, n(1),
Akaike's information criterion and Bayesian information criterion
-----------------------------------------------------------------------------
Model | Obs ll(null) ll(model) df AIC BIC
-------------+---------------------------------------------------------------
. | 1 . -7591.019 4 15190.04 15182.04
-----------------------------------------------------------------------------
Note: N=1 used in calculating BIC
. estimates store FitEmpty3,
. lrtest FitEmpty3 FitEmpty2,
Likelihood-ratio test LR chi2(1) = 1898.69
(Assumption: FitEmpty2 nested in FitEmpty3) Prob > chi2 = 0.0000
Note: The reported degrees of freedom assumes the null hypothesis is not on the boundary of the parameter space. If this is not true, then the
reported test is conservative.
Note: LR tests based on REML are valid only when the fixed-effects specification is identical for both models.
.
. display as result "Eq 12.11: Add Fixed Effects of Item Predictors"
Eq 12.11: Add Fixed Effects of Item Predictors
. mixed logrt c.rel3 c.sal3 c.rel3#c.sal3, ///
> || _all: R.subjectid , variance reml ///
> || _all: R.itemid ,
Performing EM optimization:
Performing gradient-based optimization:
Iteration 0: log restricted-likelihood = -7590.5374
Iteration 1: log restricted-likelihood = -7590.5374
Computing standard errors:
Mixed-effects REML regression Number of obs = 7646
Group variable: _all Number of groups = 1
Obs per group: min = 7646
avg = 7646.0
max = 7646
Wald chi2(3) = 19.44
Log restricted-likelihood = -7590.5374 Prob > chi2 = 0.0002
-------------------------------------------------------------------------------
logrt | Coef. Std. Err. z P>|z| [95% Conf. Interval]
--------------+----------------------------------------------------------------
rel3 | -.0502735 .0237461 -2.12 0.034 -.096815 -.003732
sal3 | -.1377454 .0421438 -3.27 0.001 -.2203457 -.0551451
|
c.rel3#c.sal3 | -.0114669 .0199695 -0.57 0.566 -.0506065 .0276727
|
_cons | 1.612585 .0569471 28.32 0.000 1.500971 1.7242
-------------------------------------------------------------------------------
------------------------------------------------------------------------------
Random-effects Parameters | Estimate Std. Err. [95% Conf. Interval]
-----------------------------+------------------------------------------------
_all: Identity |
var(R.subjec~d) | .1802798 .0215928 .1425592 .2279811
-----------------------------+------------------------------------------------
_all: Identity |
var(R.itemid) | .0941133 .0199527 .0621141 .1425973
-----------------------------+------------------------------------------------
var(Residual) | .3898702 .006391 .3775431 .4025998
------------------------------------------------------------------------------
LR test vs. linear regression: chi2(2) = 3265.05 Prob > chi2 = 0.0000
Note: LR test is conservative and provided only for reference.
. estat ic, n(1),
Akaike's information criterion and Bayesian information criterion
-----------------------------------------------------------------------------
Model | Obs ll(null) ll(model) df AIC BIC
-------------+---------------------------------------------------------------
. | 1 . -7590.537 7 15195.07 15181.07
-----------------------------------------------------------------------------
Note: N=1 used in calculating BIC
. * Multivariate Test of 3 Item Predictor Effects
. test (c.rel3=0) (c.sal3=0) (c.rel3#c.sal3=0)
( 1) [logrt]rel3 = 0
( 2) [logrt]sal3 = 0
( 3) [logrt]c.rel3#c.sal3 = 0
chi2( 3) = 19.44
Prob > chi2 = 0.0002
. estimates store FitItem,
.
. display as result "Ch 12: Remove Item Random Intercept Variance"
Ch 12: Remove Item Random Intercept Variance
. mixed logrt c.rel3 c.sal3 c.rel3#c.sal3, ///
> || _all: R.subjectid , variance reml,
Performing EM optimization:
Performing gradient-based optimization:
Iteration 0: log restricted-likelihood = -8269.5097
Iteration 1: log restricted-likelihood = -8269.5097
Computing standard errors:
Mixed-effects REML regression Number of obs = 7646
Group variable: _all Number of groups = 1
Obs per group: min = 7646
avg = 7646.0
max = 7646
Wald chi2(3) = 590.13
Log restricted-likelihood = -8269.5097 Prob > chi2 = 0.0000
-------------------------------------------------------------------------------
logrt | Coef. Std. Err. z P>|z| [95% Conf. Interval]
--------------+----------------------------------------------------------------
rel3 | -.0502263 .0043269 -11.61 0.000 -.0587068 -.0417458
sal3 | -.1377934 .0076733 -17.96 0.000 -.1528327 -.122754
|
c.rel3#c.sal3 | -.010456 .0036467 -2.87 0.004 -.0176033 -.0033087
|
_cons | 1.610471 .0349977 46.02 0.000 1.541877 1.679065
-------------------------------------------------------------------------------
------------------------------------------------------------------------------
Random-effects Parameters | Estimate Std. Err. [95% Conf. Interval]
-----------------------------+------------------------------------------------
_all: Identity |
var(R.subjec~d) | .1769577 .021418 .1395869 .2243335
-----------------------------+------------------------------------------------
var(Residual) | .4781735 .0078139 .4631011 .4937363
------------------------------------------------------------------------------
LR test vs. linear regression: chibar2(01) = 1907.10 Prob >= chibar2 = 0.0000
. estat ic, n(1),
Akaike's information criterion and Bayesian information criterion
-----------------------------------------------------------------------------
Model | Obs ll(null) ll(model) df AIC BIC
-------------+---------------------------------------------------------------
. | 1 . -8269.51 6 16551.02 16539.02
-----------------------------------------------------------------------------
Note: N=1 used in calculating BIC
. estimates store FitNoRandItem,
. lrtest FitItem FitNoRandItem,
Likelihood-ratio test LR chi2(1) = 1357.94
(Assumption: FitNoRandItem nested in FitItem) Prob > chi2 = 0.0000
Note: The reported degrees of freedom assumes the null hypothesis is not on the boundary of the parameter space. If this is not true, then the
reported test is conservative.
Note: LR tests based on REML are valid only when the fixed-effects specification is identical for both models.
.
. display as result "Eq 12.12: Add Fixed Main Effects of Subject Predictors"
Eq 12.12: Add Fixed Main Effects of Subject Predictors
. mixed logrt c.rel3 c.sal3 c.rel3#c.sal3 c.older c.yrs65, ///
> || _all: R.subjectid , variance reml ///
> || _all: R.itemid ,
Performing EM optimization:
Performing gradient-based optimization:
Iteration 0: log restricted-likelihood = -7459.9523
Iteration 1: log restricted-likelihood = -7459.9523
Computing standard errors:
Mixed-effects REML regression Number of obs = 7646
Group variable: _all Number of groups = 1
Obs per group: min = 7646
avg = 7646.0
max = 7646
Wald chi2(5) = 784.45
Log restricted-likelihood = -7459.9523 Prob > chi2 = 0.0000
-------------------------------------------------------------------------------
logrt | Coef. Std. Err. z P>|z| [95% Conf. Interval]
--------------+----------------------------------------------------------------
rel3 | -.0503054 .0237415 -2.12 0.034 -.096838 -.0037728
sal3 | -.1377449 .0421357 -3.27 0.001 -.2203294 -.0551604
|
c.rel3#c.sal3 | -.0114916 .0199657 -0.58 0.565 -.0506237 .0276404
|
older | .5899115 .0555622 10.62 0.000 .4810116 .6988114
yrs65 | .0202122 .0044056 4.59 0.000 .0115774 .028847
_cons | 1.312313 .048333 27.15 0.000 1.217582 1.407044
-------------------------------------------------------------------------------
------------------------------------------------------------------------------
Random-effects Parameters | Estimate Std. Err. [95% Conf. Interval]
-----------------------------+------------------------------------------------
_all: Identity |
var(R.subjec~d) | .0233284 .0035999 .0172398 .0315673
-----------------------------+------------------------------------------------
_all: Identity |
var(R.itemid) | .0940765 .019945 .0620896 .1425421
-----------------------------+------------------------------------------------
var(Residual) | .3898702 .006391 .3775431 .4025998
------------------------------------------------------------------------------
LR test vs. linear regression: chi2(2) = 1507.28 Prob > chi2 = 0.0000
Note: LR test is conservative and provided only for reference.
. estat ic, n(1),
Akaike's information criterion and Bayesian information criterion
-----------------------------------------------------------------------------
Model | Obs ll(null) ll(model) df AIC BIC
-------------+---------------------------------------------------------------
. | 1 . -7459.952 9 14937.9 14919.9
-----------------------------------------------------------------------------
Note: N=1 used in calculating BIC
. * Multivariate Test of 2 Subject Predictor Effects
. test (c.older=0) (c.yrs65=0)
( 1) [logrt]older = 0
( 2) [logrt]yrs65 = 0
chi2( 2) = 765.18
Prob > chi2 = 0.0000
. estimates store FitSubject,
.
. display as result "Ch 12: Remove Subject Random Intercept Variance"
Ch 12: Remove Subject Random Intercept Variance
. mixed logrt c.rel3 c.sal3 c.rel3#c.sal3 c.older c.yrs65, ///
> || _all: R.itemid , variance reml,
Performing EM optimization:
Performing gradient-based optimization:
Iteration 0: log restricted-likelihood = -7573.6119
Iteration 1: log restricted-likelihood = -7573.6119
Computing standard errors:
Mixed-effects REML regression Number of obs = 7646
Group variable: _all Number of groups = 1
Obs per group: min = 7646
avg = 7646.0
max = 7646
Wald chi2(5) = 2855.58
Log restricted-likelihood = -7573.6119 Prob > chi2 = 0.0000
-------------------------------------------------------------------------------
logrt | Coef. Std. Err. z P>|z| [95% Conf. Interval]
--------------+----------------------------------------------------------------
rel3 | -.0503109 .0237036 -2.12 0.034 -.0967692 -.0038527
sal3 | -.1373275 .0420683 -3.26 0.001 -.2197799 -.0548751
|
c.rel3#c.sal3 | -.0114482 .0199339 -0.57 0.566 -.0505178 .0276215
|
older | .5881574 .0287204 20.48 0.000 .5318664 .6444483
yrs65 | .0201377 .0022896 8.80 0.000 .0156502 .0246253
_cons | 1.312469 .0456966 28.72 0.000 1.222905 1.402032
-------------------------------------------------------------------------------
------------------------------------------------------------------------------
Random-effects Parameters | Estimate Std. Err. [95% Conf. Interval]
-----------------------------+------------------------------------------------
_all: Identity |
var(R.itemid) | .0936142 .0198811 .0617404 .1419429
-----------------------------+------------------------------------------------
var(Residual) | .4128586 .0067005 .3999325 .4262026
------------------------------------------------------------------------------
LR test vs. linear regression: chibar2(01) = 1279.96 Prob >= chibar2 = 0.0000
. estat ic, n(1),
Akaike's information criterion and Bayesian information criterion
-----------------------------------------------------------------------------
Model | Obs ll(null) ll(model) df AIC BIC
-------------+---------------------------------------------------------------
. | 1 . -7573.612 8 15163.22 15147.22
-----------------------------------------------------------------------------
Note: N=1 used in calculating BIC
. estimates store FitNoRandSubject,
. lrtest FitSubject FitNoRandSubject,
Likelihood-ratio test LR chi2(1) = 227.32
(Assumption: FitNoRandSub~t nested in FitSubject) Prob > chi2 = 0.0000
Note: The reported degrees of freedom assumes the null hypothesis is not on the boundary of the parameter space. If this is not true, then the
reported test is conservative.
Note: LR tests based on REML are valid only when the fixed-effects specification is identical for both models.
.
. ****** END CHAPTER 12 MODELS ******
.
. * Close log
. log close STATA_Chapter12
name: STATA_Chapter12
log: C:\Dropbox\PilesOfVariance\Chapter12\STATA\STATA_Chapter12_Output.smcl
log type: smcl
closed on: 21 Oct 2014, 10:33:39
------------------------------------------------------------------------------------------------------------------------------------------------------