G89.2247 Lect 2 1 G89.2247 Lecture 2 • Regression as paths and covariance structure • Alternative “saturated” path models • Using matrix notation to write linear models • Multivariate Expectations • Mediation
G89.2247 Lect 2 1
G89.2247Lecture 2
• Regression as paths and covariance structure
• Alternative “saturated” path models
• Using matrix notation to write linear models
• Multivariate Expectations
• Mediation
G89.2247 Lect 2 2
Question: Does exposure to childhood foster care (X) lead to adverse outcomes (Y) ?
• Example of purported "causal model" X Y
Y = B0 + B1X + e• Regression approach
B0 and B1 can be estimated using OLS Estimates depend on sample standard deviations of Y and X, sample means,
and covariance between Y and X• B1 = SXY/S2
X
• B0 = MY -B1MX
Correlation, rXY = SXY/SXSY, can be used to estimate the variance of the residual, e, V(e).
• S2e = S2
Y(1-r2XY) = S2
Y - S2XY/S2
X
eB1
G89.2247 Lect 2 3
A Covariance Structure Approach
• If we have data on Y and X we can compute a covariance matrix
• This estimates the population covariance structure,
Y can itself be expressed as B2
12X + 2
e
Three statistics in the sample covariance matrix are available to estimate three population parameters
2
2
XXY
YXY
SS
SSS
221
2
2`
2
XX
Y
XXY
YXY
B
G89.2247 Lect 2 4
Covariance Structure Approach, Continued
• A structural model that has the same number of parameters as unique elements in the covariance matrix is "saturated".
• Saturated models always fit the sample covariance matrix.
221
21
2221
XX
XeX
B
BB
G89.2247 Lect 2 5
Another saturated model: Two explanatory variables
• The first model is likely not to yield an unbiased estimate of foster care because of selection factors (Isolation failure).
• Suppose we have a measure of family disorganization (Z) that is known to have an independent effect on Y and also to be related to who is assigned to foster care (X)
YX
Z
e
XZ
G89.2247 Lect 2 6
Covariance Structure Expression
• The model: Y=b0+b1X+b2Z+e If we assume E(X)=E(Z)=E(Y)=0 and V(X) = V(Z) = V(Y) = 1 then b0=0 and 's are standardized
• The parameters can be expressed
When sample correlations are substituted, these expressions give the OLS estimates of the regression coefficients.
211 XZ
XZYZYX
22
1 XZ
XZYXYZ
G89.2247 Lect 2 7
Covariance Structure: 2 Explanatory Variables
• In the standardized case the covariance structure is:
• Each correlation is accounted by two components, one direct and one indirect
• There are three regression parameters and three covariances.
1
1
1
1
1
1
12
21
XZXZ
XZ
XZZY
XZXY
ZYXY
G89.2247 Lect 2 8
The more general covariance matrix for two IV multiple regression
• If we do not assume variances of unity the regression model implies
212
221
2
12
21
1
1
1
ZYXZZYXZZ
XYXZX
Y
XZXZ
XZ DD
Z
X
Y
V
G89.2247 Lect 2 9
More Math Review for SEM
• Matrix notation is useful
Z
X
Y
XZXZ
XZXZ
XZXZ
Z
X
Y
Z
X
Y
V
00
00
00
1
1
1
00
00
00
12
21
1221
212
221
2
ZYXZZYXZZ
XYXZX
Y
G89.2247 Lect 2 10
A Matrix Derivation of OLS Regression
• OLS regression estimates make the sum of squared residuals as small as possible.If Model is
Then we choose B so that e'e is minimized.
• The minimum will occur when the residual vector is orthogonal to the regression planeIn that case, X'e = 0
eBXY
G89.2247 Lect 2 11
When will X'e = 0? When e is the residual from an OLS fit.
BXXYXBXYXeX ˆ''ˆ'0'
YXXXB
YXXXBXXXX
YXBXX
''ˆ
''ˆ''
'ˆ'
1
11
G89.2247 Lect 2 12
Multivariate Expectations
• There are simple multivariate generalizations of the expectation facts:E(X+k) = E(X)+k = x+kE(k*X) = k*E(X) = k*x
V(X+k) = V(X) = x2
V(k*X) = k2*V(X) = k2*x2
• Let XT=[X1 X2 X3 X4], T=[] and let k be scalar valueE(k*X) = k*E(X) = k*E(X+k* 1) = {E(X) + k* 1} = + k*1
G89.2247 Lect 2 13
Multivariate Expectations
• In the multivariate case Var(X) is a matrixV(X)=E[(X-) (X-)T]
24434241
34233231
24232221
14131221
G89.2247 Lect 2 14
Multivariate Expectations
• The multivariate generalizations of V(X+k) = V(X) = x
2
V(k*X) = k2*V(X) = k2*x2
• Are: Var(X + k*1) = Var(k* X) = k2
• Let cT = [c1 c2 c3 c4]; cT X is a linear combination of the X's. Var(cT X) = cT c
• This is a scalar value• If this positive for all values of c then is positive
definite
G89.2247 Lect 2 15
Semi Partial Regression Adjustment
• The multiple regression coefficients are estimated taking all variables into accountThe model assumes that for fixed X, Z has an effect
of magnitude Z.Sometimes people say "controlling for X"
• The model explicitly notes that Z has two kinds of association with YA direct association through Z (X fixed)An indirect association through X (magnitude
XXZ)
G89.2247 Lect 2 16
Pondering Model 1:Simple Multiple Regression
• The semi-partial regression coefficients are often different from the bivariate correlationsAdjustment effectsSuppression effects
• Randomization makes XZ = 0 in probability.
YX
Z
e
XZ
G89.2247 Lect 2 17
Mathematically Equivalent Saturated Models
• Two variations of the first model suggest that the correlation between X and Z can itself be represented structurally.
YX
Z
eY
eZ
YX
Z
eY
eX
G89.2247 Lect 2 18
Representation of Covariance Matrix
• Both models imply the same correlation structure
• The interpretation, however, is very different.
1
1
1
1
1
1
3132
231
XZZY
XZXY
ZYXY
G89.2247 Lect 2 19
Model 2:X leads to Z and Y
• X is assumed to be causally prior to Z.The association between X and Z is due to X effects.
• Z partially mediates the overall effect of X on YX has a direct effect 1 on YX has an indirect effect on Y through ZPart of the bivariate association between Z and Y is spurious (due
to common cause X)
YX
Z
eY
eZ
G89.2247 Lect 2 20
Model 3:Z leads to X and Y
• Z is assumed to be causally prior to X.The association between X and Z is due to Z effects.
• X partially mediates the overall effect of Z on YZ has a direct effect 2 on YZ has an indirect effect on Y through XPart of the bivariate association between X and Y is spurious (due
to common cause Z)
YX
Z
eY
eX
G89.2247 Lect 2 21
Choosing between models
• Often authors claim a model is good because it fits to data (sample covariance matrix)All of these models fit the same (perfectly!)
• Logic and theory must establish causal order• There are other possibilities besides 2 and 3
In some instances, X and Z are dynamic variables that are simultaneously affecting each other
In other instances both X and Z are outcomes of an additional variable, not shown.
G89.2247 Lect 2 22
Mediation: A theory approach
• Sometimes it is possible to argue on theoretical grounds thatZ is prior to X and YX is prior to YThe effect of Z on Y is completely accounted for by the
indirect path through X.
• This is an example of total mediation• If is fixed to zero, then Model 3 is no longer
saturated.Question of fit becomes informativeTotal mediation requires strong theory
G89.2247 Lect 2 23
A Flawed Example
• Someone might try to argue for total mediation of family disorganization on low self-esteem through placement in foster care
• Baron and Kenny(1986) criteria might be metZ is significantly related to YZ is significantly related to XWhen Y is regressed on Z and X, is significant but is not
significant.
• Statistical significance is a function of sample size.• Logic suggests that children not assigned to foster care
who live in a disorganized family may suffer directly.
G89.2247 Lect 2 24
A More Compelling Example of Complete Mediation
• If Z is an experimentally manipulated variable such as a prime
• X is a measured process variable• Y is an outcome logically subsequent to X
It should make sense that X affects Y for all levels of Z
E.g. Chen and Bargh (1997)• Are participants who have been subliminally primed
with negative stereotype words more likely to have partners who interact with them in a hostile manner?