Newsom 1

USP 656 Multilevel Regression

Winter 2006
xvxgdfgdfgdjgg (i.e. 0 mg 5 mg !nd "0 mg#. $owever if t%e rese!r&%er w!nts to m!'e inferen&es
(e)ond t%e *!rti&ul!r v!lues of t%e &g%dfdgfgdfgsfgsdfgsdfgdffffffffff'''''dfg%dg%(le used in t%e stud)
! r!ndom effe&ts model is used. + &ommon ex!m*le would (e t%e use of *u(li& !rt wor's
re*resenting low moder!te !nd &v(&(&(&v(&v(
Fixed and Random Coefficients in Multilevel Regression
,%e r!ndom vs. fixed d%gdfgdfgdfgdfgdfgdfgdf'!nd effe&ts is im*ort!nt in multilevel regression. -n
multilevel regression models (ot% level." &v&v(&v(v&( to v!r) r!ndoml) !&ross grou*s. /e&!use of
t%e !ssum*tions !(out t%eir error distri(utions we &!ll t%eir v!ri!n&es oo=var(U0j) !nd 11=var(U1j),
0r!ndom &oeffi&ients.1
,%is me!ns t%!t we &!n t%in' !(out 0j !nd 1j !s !'in to t%e r!ndom variables - des&ri(ed
!(ove. -nste!d of !ttem*ting to gener!li2e (e)ond t%e *!rti&ul!r v!lues of t%e inde*endent v!ri!(le
we !re !ttem*ting to gener!li2e (e)ond t%e *!rti&ul!r grou*s in t%e stud). 3or inst!n&e we m!) %!ve
50 &om*!nies (ut we wis% to gener!li2e to ! l!rger universe of &om*!nies w%en we ex!mine t%e
me!ns (inter&e*ts# or t%e x.) rel!tions%i* (slo*es#.
-n $4M v!ri!n&es for t%e inter&e*ts !nd slo*es !re estim!ted () def!ult. ,%!t is w%en )ou
first &onstru&t ! model U0j !nd U1j !re estim!ted () def!ult. $owever t%e rese!r&%er %!s t%e &%oi&e
of setting U0j !nd U1j to (e 2ero. /) setting t%em to 2ero we !re testing ! model in w%i&% we !ssume
0j !nd 1j do not v!r) r!ndoml) !&ross grou*s. -n f!&t t%eir v!ri!n&e is !ssumed to (e 2ero so t%e)
!re !ssumed to (e &onst!nt or 0nonv!r)ing1 !&ross grou*s. 3or ex!m*le fixed nonv!r)ing inter&e*ts
would im*l) t%e grou* !ver!ge for t%e de*endent v!ri!(le is !ssumed to (e e5u!l in e!&% grou*.
+lt%oug% we t)*i&!ll) refer to t%is &onstr!int !s 0fixing t%e inter&e*ts1 or 0fixing t%e slo*es1 t%e term is
somew%!t loosel) !**lied (e&!use we !re re!ll) !ssuming t%e) !re fixed and nonvarying.
Intercept Model and Random Effects ANOVA
-n t%e sim*lest $4M model t%ere !re no *redi&tors. We %!ve one level." e5u!tion !nd one level.2
e5u!tion6
0 ij j ij
Y R = +
w%ere Y is t%e de*endent v!ri!(le 0j is t%e inter&e*t !nd Rij is t%e residu!l or error. 0j &!n (e
t%oug%t of !s t%e me!n of e!&% grou*. ,%e level.2 e5u!tion !lso %!s no *redi&tors in it7s sim*lest
form6
0 00 0 j j
U = +
w%ere
0j
is t%e de*endent v!ri!(le
00
is t%e level.2 inter&e*t !nd U
0j
is t%e level.2 error. -n t%is
e5u!tion 00 re*resents t%e gr!nd me!n or t%e me!n of t%e inter&e*ts. U0j re*resents t%e devi!tion of
e!&% me!n from t%e gr!nd me!n. W%en t%e !ver!ge devi!tion is l!rge t%ere !re l!rge grou*
differen&es.
-t &!n (e s%own t%!t t%is is re!ll) t%e +N89+ model. +N89+ ex!mines t%e devi!tions of grou* me!ns
from t%e gr!nd me!n. /e&!use we !ssume t%!t t%e grou* me!ns re*resented () 0j !nd t%us t%eir
devi!tions !re r!ndoml) v!r)ing t%is model is e5uiv!lent to t%e random effects +N89+ model. We
&!n rewrite our two $4M e5u!tions !s ! single e5u!tion if we *lug in t%e level.2 terms into level."
e5u!tion.
00 0 ij j ij
Y U R = + +
-n t%is e5u!tion 00 re*resents t%e gr!nd me!n t%e U0j term re*resents t%e devi!tions from t%e me!n
(or me!n differen&es or tre!tment effe&t if )ou would li'e# !nd Rij re*resents t%e wit%in.grou* error or
v!ri!tion. -n more m!t%em!ti&!ll) oriented +N89+ text(oo's )ou will see !n +N89+ model written
somet%ing li'e6
Newsom 2
USP 656 Multilevel Regression
Winter 2006
ij j ij
Y e = + +
-f t%e model is !n r!ndom effe&ts +N89+ j is !ssumed to %!ve ! r!ndom distri(ution. -n multilevel
regression (e&!use we !ssume U0j to v!r) r!ndoml) t%e sim*le $4M model wit% no level." or level.2
*redi&tors is e5uiv!lent to t%e r!ndom effe&ts +N89+ model.
W%en we !dd *redi&tors to t%e level." e5u!tion t%e) !re &ov!ri!tes !nd t%e model (e&omes !
r!ndom effe&ts +N:89+ in w%i&% t%e me!ns !re !djusted for t%e &ov!ri!te.
Newsom 3
USP 656 Multilevel Regression
Winter 2006
Summary a!le
Random vs"
Fixed
#efinition Example $se in Multilevel
Regression
Varia!les Random variable6 ("# is !ssumed to (e me!sured wit%
me!surement error. ,%e s&ores !re ! fun&tion of ! true
s&ore !nd r!ndom error; (2# t%e v!lues &ome from !nd
!re intended to gener!li2e to ! mu&% l!rger *o*ul!tion
of *ossi(le v!lues wit% ! &ert!in *ro(!(ilit) distri(ution
(e.g. norm!l distri(ution#; (<# t%e num(er of v!lues in
t%e stud) is sm!ll rel!tive to t%e v!lues of t%e v!ri!(le !s
it !**e!rs in t%e *o*ul!tion it is dr!wn from. Fixed
variable6 ("# !ssumed to (e me!sured wit%out
me!surement error; (2# desired gener!li2!tion to
*o*ul!tion or ot%er studies is to t%e s!me v!lues; (<#
t%e v!ri!(le used in t%e stud) &ont!ins !ll or most of t%e
v!ri!(le7s v!lues in t%e *o*ul!tion.
-t is im*ort!nt to distinguis% (etween ! v!ri!(le t%!t is
varying !nd ! v!ri!(le t%!t is random. + fixed v!ri!(le
&!n %!ve different v!lues it is not ne&ess!ril) inv!ri!nt
(e5u!l# !&ross grou*s.
Random variable:
*%otogr!*%s
re*resenting individu!ls
wit% differing levels of
!ttr!&tiveness
m!ni*ul!ted in !n
ex*eriment &ensus
tr!&'s
Fixed variable:
gender r!&e or
intervention vs. &ontrol
grou*.
Predi&tor
v!ri!(les in M4R
gener!ll)
!ssumed to (e
fixed
Effects Random effect6 ("# different st!tisti&!l model of
regression or +N89+ model w%i&% !ssumes t%!t !n
inde*endent v!ri!(le is r!ndom; (2# gener!ll) used if
t%e levels of t%e inde*endent v!ri!(le !re t%oug%t to (e
! sm!ll su(set of t%e *ossi(le v!lues w%i&% one wis%es
to gener!li2e to; (<# will *ro(!(l) *rodu&e l!rger
st!nd!rd errors (less *owerful#. Fixed effect6 ("#
st!tisti&!l model t)*i&!ll) used in regression !nd +N89+
!ssuming inde*endent v!ri!(le is fixed; (2#
gener!li2!tion of t%e results !**l) to simil!r v!lues of
inde*endent v!ri!(le in t%e *o*ul!tion or in ot%er
studies; (<# will *ro(!(l) *rodu&e sm!ller st!nd!rd
errors (more *owerful#.
Random effect6
r!ndom effe&ts +N89+
r!ndom effe&ts
regression
Fixed effect6 fixed
effe&ts +N89+ fixed
effe&ts regression
-nter&e*t onl)
models in M4R
!re e5uiv!lent to
r!ndom effe&ts
+N89+ or
+N:89+.
Coefficients Random coefficient: term !**lies onl) to M4R
!n!l)ses in w%i&% inter&e*ts slo*es !nd v!ri!n&es &!n
(e !ssumed to (e r!ndom. M4R !n!l)ses most
t)*i&!ll) !ssume r!ndom &oeffi&ients. 8ne &!n
&on&e*tu!li2e t%e &oeffi&ients o(t!ined from t%e level."
regressions !s ! t)*e of r!ndom v!ri!(le w%i&% &omes
from !nd gener!li2es to ! distri(ution of *ossi(le v!lues.
=rou*s !re &on&eived of !s ! su(set of t%e *ossi(le
grou*s.
Fixed coefficient: ! &oeffi&ient &!n (e fixed to (e non.
v!r)ing (inv!ri!nt# !&ross grou*s () setting its (etween
grou* v!ri!n&e to 2ero.
R!ndom &oeffi&ients must (e v!ri!(le !&ross grou*s.
:on&e*tu!ll) fixed &oeffi&ients m!) (e inv!ri!nt or
v!r)ing !&ross grou*s.
Random coefficient6
t%e level.2 *redi&tor
!ver!ge in&ome is
used to *redi&t s&%ool
*erform!n&e in e!&%
s&%ool. -nter&e*t
v!lues for s&%ool
*erform!n&e !re
!ssumed to (e !
s!m*le of t%e inter&e*ts
from ! l!rger *o*ul!tion
of s&%ools.
Fixed coefficient6
slo*es or inter&e*ts
&onstr!ined to (e e5u!l
over different s&%ools.
/ot% used in
M4R. Slo*es !nd
inter&e*t v!lues
&!n (e
&onsidered to (e
fixed or r!ndom
de*ending on
rese!r&%ers
!ssum*tions !nd
%ow t%e model is
s*e&ified. ,%e
v!ri!n&e of t%e
slo*es or
inter&e*ts !re
&onsidered
r!ndom
&oeffi&ients.