Modeling Continuous

Longitudinal Data
Introduction to continuous
longitudinal data:
Examples
 Homeopathy vs. placebo in
treating pain after surgery

Day of surgery

Mean pain
assessments by
visual analogue
scales (VAS)

Days 1-7 after surgery

(morning and evening)

Copyright ©1995 BMJ Publishing Group Ltd.

Lokken, P. et al. BMJ 1995;310:1439-1442
 Divalproex vs. placebo for
treating bipolar depression

Davis et al. “Divalproex in the treatment of bipolar depression: A placebo controlled study.” J
Affective Disorders 85 (2005) 259-266.
 Randomized trial of in-field
treatments
of acute mountain sickness

Mean (SD) score of acute

mountain sickness in subjects
treated with simulated descent
(One hour of treatment in the
hyperbaric chamber) or
dexamethasone.

Copyright ©1995 BMJ Publishing Group Ltd. Keller, H.-R. et al. BMJ 1995;310:1232-1235
 Pint of milk vs. control on
bone acquisition in
adolescent females

Mean (SE) percentage increases in

total body bone mineral and bone
P values are density over 18 months.
for the
differences
between
groups by
repeated
measures
analysis of
variance

Copyright ©1997 BMJ Publishing Group Ltd. Cadogan, J. et al. BMJ 1997;315:1255-1260
 Counseling vs. control on
smoking in pregnancy

Copyright ©2000 BMJ Publishing Group Ltd.

Hovell, M. F et al. BMJ 2000;321:337-342
 Longitudinal data: broad
form
id time1 time2 time3
time4

1 31 29 15 26
2 24 28 20 32
3 14 20 28 30
4 38 34 30 34
5 25 29 25 29
6 30 28 16 34

Hypothetical data from Twisk, chapter 3, page 26,

table
W. R. Twisk. 3.4
Applied Longitudinal Data Analysis for Epidemiology: A Practical Guide. Cambridge University Press, 2003.
 Longitudinal data: Long
form
id time id time
score score
1 1 31 4 1 38
1 2 29
Hypothetical
data from 1 3 15
Twisk, chapter
3, page 26,
4 2 34
table 3.4 4 3 30
1 4 26 4 4 34
2 1 24 5 1 25
2 2 28 5 2 29
2 3 20 5 3 25
2 4 32 5 4 29
3 1 14 6 1 30
Converting data from broad
to long in SAS…

data long;
set broad;
time=1; score=time1; output;
time=2; score=time2; output;
time=3; score=time3; output;
time=4; score=time4; output;
run;
Profile plots (use long
form)

The plot tells a lot!

Mean response plot
Superimposed…
smoothed
smoothed
Superimposed…
 Two groups (e.g., treatment
placebo)

id group time1 time2 time3

time4

1 A 31 29 15 26
2 A 24 28 20 32
3 A 14 20 28 30
4 B 38 34 30 34
5 B 25 29 25 29
6 B 30 28 16 34

Hypothetical data from Twisk, chapter 3, page 40,

table 3.7
Profile plots by group

A
Mean plots by group

A
 Possible questions…
 Overall, are there significant differences between time
points?
 From plots: looks like some differences (time3 and 4 look
different)
 Overall, are there significant changes from baseline?
 From plots: at time3 or time4 maybe
 Do the two groups differ at any time points?
 From plots: certainly at baseline; some difference everywhere
 Do the two groups differ in their responses over time?**
 From plots: their response profile looks similar over time,
though A and B are closer by the end.
 Statistical analysis
strategies
 Strategy 1: ANCOVA on the final
measurement, adjusting for baseline
differences (end-point analysis) Traditional
 Strategy 2: repeated-measures ANOVA approaches:
“Univariate” approach this week
 Strategy 3: “Multivariate” ANOVA approach

 Strategy 4: GEE Newer

 Strategy 5: Mixed Models approaches:
next week
 Strategy 6: Modeling change
In two/three weeks
 Comparison of traditional
and new methods

FROM:
Ralitza Gueorguieva, PhD; John H. Krystal, MD Move Over ANOVA : Progress in Analyzing Repeated-Measures Data and Its Reflection in Papers Published in the Archives
of General Psychiatry. Arch Gen Psychiatry. 2004;61:310-317.
 Things to consider:
1. Spacing of time intervals

Repeated-measures ANOVA and MANOVA require that all subjects
measured at same time intervals—our plots above assumed this too!

MANOVA weights all time intervals evenly (as if evenly spaced)
2. Assumptions of the model

ALL strategies assume normally distributed outcome and
homogeneity of variances

But all strategies are robust against this assumption,
especially if data set is >30

**Univariate repeated-measures ANOVA assumes sphericity, or
compound symmetry
3. Missing Data

All traditional analyses require imputation of missing data

(also need to know: does the SAS PROC require long or broad form of
data?)
 Compound symmetry
Compound symmetry requires :

(a) The variances of the outcome variable

must be the same at each time point

(b) The correlation between repeated

measurements are equal, regardless of
the time interval between
measurements.
 (a) Variances at each time
points (visually)

Does variance look equal across time points??

--Looks like most variability at time1 and least at
time4…
 (a) Variances at each time
points (numerically)
id time1 time2 time3
time4

1 31 29 15 26
2 24 28 20 32
3 14 20 28 30
4 38 34 30 34
5 25 29 25 29
6 30 28 16 34
Variance: 9.76667
65.60000 20.40000 39.46667
(b) Correlation (covariance)
across time points
time1 time2 time3 time4

time1 1.00000 0.94035 -0.14150

0.28445

time2 0.94035 1.00000 -0.02819

0.26921

time3 -0.14150 -0.02819 1.00000

0.27844

time4 0.28445
Certainly do NOT0.26921 0.27844
have equal
1.00000
correlations!
Time1 and time2 are highly correlated,
 Compound symmetry would
look like…

time1 time2 time3 time4

time1 1.00000 -0.04878 -0.04878 -0.04878

time2 -0.04878 1.00000 -0.04878 -0.04878

time3 -0.04878 -0.04878 1.00000 -0.04878

time4 -0.04878 -0.04878 -0.04878 1.00000

Missing Data
 Very important to fill in missing data!
Otherwise, you have to throw out the whole
observation.
 With missing data, changes in the mean over
time may just reflect drop-out pattern; you
cannot compare time point 1 with 50 people
to time point 2 with 35 people!
 We will implement classic “last observation
carried forward” strategy for simplicity
 Other more complicated imputation
strategies may be more appropriate
 LOCF

Subject HRSD 1 HRSD 2 HRSD 3 HRSD 4

Subject 20 13
1
Subject 21 21 20 19
2
Subject 19 18 10 6
3
Subject 30 25 23
4
 LOCF
Last Observation Carried Forward
Subject HRSD 1 HRSD 2 HRSD 3 HRSD 4
Subject 20 13 13 13
1
Subject 21 21 20 19
2
Subject 19 18 10 6
3
Subject 30 30 25 23
4
 Strategy 1: End-point
analysis
Removes repeated measures problem by
considering only a single time point (the final
one).
Ignores intermediate data completely
Asks whether or not the two group means differ at
the final time point, adjusting for differences at
baseline (using ANCOVA).
proc glm data=broad;
class group;
model time4 = time1 group;
run;

Comparing groups at every follow-up time point in this

way would hugely increase your type I error.
 Strategy 1: End-point
analysis
Sum of
Source DF Squares Mean Square F Value Pr > F

Model 2 13.50000000 6.75000000 0.57 0.6155

Error 3 35.33333333 11.77777778

Corrected Total 5 48.83333333

R-Square Coeff Var Root MSE time4 Mean

0.276451 11.13041 3.431877 30.83333

Source DF Type I SS Mean Square F Value Pr > F

time1 1 3.95121951 3.95121951 0.34 0.6031

group 1 9.54878049 9.54878049 0.81 0.4343

group time4 LSMEAN Pr > |t|

A 29.3333333 0.4343
B 32.3333
 Strategy 1: End-point
analysis
Sum of
Source DF Squares Mean Square F Value Pr > F

Model 2 13.50000000 6.75000000 0.57 0.6155

Error 3 35.33333333 11.77777778

Corrected Total 5 48.83333333 Least-squares

means of the two
R-Square Coeff Var Root MSE time4 Mean
groups at time4,
0.276451 11.13041 3.431877 adjusted
30.83333 for baseline
differences (not
Source DF Type I SS Mean Square F Value Pr > F
significantly
time1
group
1
1
3.95121951
9.54878049
different)
3.95121951
9.54878049
0.34
0.81
0.6031
0.4343

group time4 LSMEAN Pr > |t|

A 29.3333333 0.4343
B 32.3333
 From end-point analysis…
 Overall, are there significant differences between
time points?
 Can’t say
 Overall, are there significant changes from
baseline?
 Can’t say
 Do the two groups differ at any time points?
 They don’t differ at time4
 Do the two groups differ in their responses over
time?
 Can’t say
Strategy 2: univariate
repeated measures ANOVA
(rANOVA)
Just good-old regular ANOVA, but
accounting for between subject differences
 BUT first… Naive analysis
 Run ANOVA on long form of data,
ignoring correlations within subjects (also
ignoring group for now):

proc anova data=long;

class time;
model score= time ;
run;
Compares means from each time point as if they were
independent samples. (analogous to using a two-
sample t-test when a paired t-test is appropriate).
 One-way ANOVA (naïve)
id time1 time2 time3
time4 MEAN
Within time

1 31 29 15 26
2 24 28 20 32Between
3 14 20 28 30times
4 38 34 30 34
5 25 29 25 29
6 30 28 16 34
MEAN: 27.00 28.00 22.33 30.83
2 2 2 2
SSB (between t imes)
27.006 x[( 27  27 )  ( 28  27 )  ( 22 . 33  27 )  (30 . 83  27 ) ] 224.79

SSW (within time) (31  27) 2  (24  27) 2  .....  (29  30.83) 2  (34  30.83) 2 676.17
 One-way ANOVA results
The ANOVA Procedure
Dependent Variable: score
Sum of
Source DF Squares Mean Square F Value Pr > F
Model 3 224.7916667 74.9305556 2.22 0.1177
Error 20 676.1666667 33.8083333
Corrected Total 23 900.9583333

Source DF Anova SS Mean Square F Value Pr > F

time 3 224.7916667 74.9305556 2.22 0.1177

Twisk: Output 3.3

 Univariate repeated-
measures ANOVA

Explain away some error variability by accounting for

differences between subjects:
-SSE was 676.17
-This will be reduced by variability between subjects
proc glm data=broad;
model time1-time4=;
repeated time;
run; quit;
 rANOVA Between
id time1 time2 time3 time4 MEAN
subjects
1 31 29 15 26 25.25
2 24 28 20 32 26.00
3 14 20 28 30 23.00
4 38 34 30 34 34.00
5 25 29 25 29 27.00
6 30 28 16 34 27.00
MEAN: 27.00 28.00 22.33 30.83 27.00
SSB (between times) 224.79 (from before)
SSid (between subjects) 4 x[(25.25  27) 2  (26  27) 2  (23  27) 2  ...  (27  27) 2 ] 276.21

unexplained variability 676.17 - 276.21 399.96

 Idea of G-G and H-F corrections, analogous to pooled vs.
unpooled variance ttest: if we have to estimate more
things because variances/covariances aren’t equal, then
we lose some degrees of freedom and p-value increases.

rANOVA results Repeated measures p-value = .0752

After G-G correction for non-
sphericity=.1311
(H-F correction gives .1114)
The GLM Procedure
Repeated Measures Analysis of Variance
Univariate Tests of Hypotheses for Within Subject Effects

Adj Pr > F
Source DF Type III SS Mean Square F Value Pr > F G - G H - F

time 3 224.7916667 74.9305556 2.81 0.0752 0.1311 0.1114

Error(time) 15 399.9583333 26.6638889
These epsilons
Greenhouse-Geisser Epsilon 0.4857 should be 1.0 if
Huynh-Feldt Epsilon 0.6343 sphericity holds.
Sphericity
assumption
appears violated.

Between time
Unexplained variability
variability
 With two groups: Naive
analysis
 Run ANOVA on long form of data,
ignoring correlations within subjects:

proc anova data=long;

class time;
model score= time group group*time;
run;
As if there are 8 independent samples: 2 groups at
each time point.
 Two-way ANOVA (naïve)
grp time1 time2 time3 time4 MEAN
A 31 29 15Within
26time
A 24 28 20 32
A 14 20 28 30
MEAN: 23.00 25.67 21.00 19.33 24.75 Overall
Within time mean=27
B 38 34 30 34
B 25 29 25 29 Between
B 30 28 16 34 groups
MEAN: 31.00 30.33 23.67 32.33 29.33
SSB(between times) 224.79 (from before)
SSB (between groups) 12 x[(29.33  27) 2  (24.75  27) 2 ] 126.04
SSE [( 31  23) 2  ( 24  23) 2  (14  23) 2  ...  ( 29  25.67 )] 523.33
Recall: SST=900.9583333; group by time=900.9583-523.33-224.79-126.04=26.79
 Results: Naïve analysis
The ANOVA Procedure

Dependent Variable: score

Sum of
Source DF Squares Mean Square F Value Pr > F

Model 7 377.6250000 53.9464286 1.65 0.1924

Error 16 523.3333333 32.7083333

Corrected Total 23 900.9583333

Source DF Anova SS Mean Square F Value Pr > F

time 3 224.7916667 74.9305556 2.29 0.1173

group 1 126.0416667 126.0416667 3.85 0.0673
time*group 3 26.7916667 8.9305556 0.27 0.8439
 Univariate repeated-
measures ANOVA

Reduce error variability by between subject differences:

-SSE was 523.33
-This will be reduced by variability between subjects

proc glm data=broad;

class group;
model time1-time4= group;
repeated time;
run; quit;
 rANOVA
grp time1 time2 time3 time4 MEANBetween
subjects in
A 31 29 15 26 25.25each group
A 24 28 20 32 26.00
A 14 20 28 30 23.00
MEAN: 23.00 25.67 21.00 19.33 24.75 Overall
mean=27
Between
B 38 34 30 34 34.00subjects in
B 25 29 25 29 27.00each group
B 30 28 16 34 27.00
MEAN: 31.00 30.33 23.67 32.33 29.33
SS id (between subjects) 4 x[( 25.25  24.75) 2  ( 26  24.75) 2  ...  ( 27  29.33) 2 ] 150.16

unexplained variability 523.33  150.17 373.167

 rANOVA results (two
groups)
The GLM Procedure Usually of
Repeated Measures Analysis of Variance
Tests of Hypotheses for Between Subjects Effects less interest!
Source DF Type III SS Mean Square F Value Pr > F

group 1 126.0416667 126.0416667 3.36 0.1408

Error 4 150.1666667 37.5416667

The GLM Procedure What we care

Repeated Measures Analysis of Variance
Univariate Tests of Hypotheses for Within Subject Effects about!
Adj Pr > F
Source DF Type III SS Mean Square F Value Pr > F G - G H - F

time 3 224.7916667 74.9305556 2.41 0.1178 0.1743 0.1283

time*group 3 26.7916667 8.9305556 0.29 0.8338 0.6954 0.8118
Error(time) 12 373.1666667 31.0972222

No apparent difference in
Greenhouse-Geisser Epsilon 0.4863
Huynh-Feldt Epsilon 0.885 responses over time
between the groups.
 From rANOVA analysis…
 Overall, are there significant differences between
time points?
 No, Time not statistically significant (p=.1743, G-G)
 Overall, are there significant changes from
baseline?
 No, Time not statistically significant
 Do the two groups differ at any time points?
 No, Group not statistically significant (p=.1408)
 Do the two groups differ in their responses over
time?**
 No, not even close; Group*Time (p-value>.60)
Strategy 3: rMANOVA
 Multivariate: More than one
dependent variable
 Multivariate Approach to repeated
measures--Treats response
variable as a multivariate response
vector.
 Not just for repeated measures,
but appropriate for other situations
with multiple dependent variables.
 Analogous to paired t-test
n

 Recall: paired t- y 2  y1
Ydiff  i 1
test: n
Ydiff
~ Tn  1
SD(Ydiff )

Paired t-test compares the difference values between

two time points to their standard error.

MANOVA is just a paired t-test where the outcome

variable is a vector of difference rather than a single
difference: N  T 1 2
Where T is the F ( )H
( N  1)(T  1)
number of time Ny y2
T Called: Hotelling's Trace
diff diff
H 
points: S 2
diff
T-1 differences
id group diff1 diff2 diff3

1 A -2 -14 11
2 A 4 -8 12
3 A 6 8 2
4 B -4 -4 4
5 B 4 -4 4
6 B -2 -12 18

Note: weights all differences equally, so hard to interpret if time

intervals are unevenly spaced.
Note: assumes differences follow a multivariate normal
distribution + multivariate homogeneity of variances assumption
On same output as
rANOVA
proc glm data=broad;
model time1-time4=;
repeated time;
run; quit;

Null hypothesis: diff1=0, diff2=0,

diff3=0
 Results (time only)
MANOVA Test Criteria and Exact F Statistics for the Hypothesis of no time Effect
H = Type III SSCP Matrix for time
E = Error SSCP Matrix

S=1 M=0.5 N=0.5

Statistic Value F Value Num DF Den DF Pr > F

Wilks' Lambda 0.24281920 3.12 3 3 0.1876

Pillai's Trace 0.75718080 3.12 3 3 0.1876
Hotelling-Lawley Trace 3.11829053 3.12 3 3 0.1876
Roy's Greatest Root 3.11829053 3.12 3 3 0.1876

•4 separate F-statistics (slightly different versions of Use Wilks’

Lambda in
MANOVA statistic) general.
Use Pillai’s Trace
•all give the same answer: change over time is not for small sample
significant sizes (when
assumptions of
•compare to rANOVA results: G-G time p-value=.13 model are
On same output as
rANOVA
proc glm data=broad;
class group;
model time1-time4= group;
repeated time;
run; quit;
 Results (two groups)
The GLM Procedure Repeated Measures Analysis of Variance
MANOVA Test Criteria and Exact F Statistics for the Hypothesis of no time Effect

Statistic Value F Value Num DF Den DF Pr > F

Wilks' Lambda 0.23333404 2.19 3 2 0.3287

Pillai's Trace 0.76666596 2.19 3 2 0.3287
Hotelling-Lawley Trace 3.28570126 2.19 3 2 0.3287
Roy's Greatest Root 3.28570126 2.19 3 2 0.3287

MANOVA Test Criteria and Exact F Statistics for the Hypothesis of no time*group Effect

Statistic Value F Value Num DF Den DF Pr > F

Wilks' Lambda 0.77496006 0.19 3 2 0.8932

Pillai's Trace 0.22503994 0.19 3 2 0.8932
Hotelling-Lawley Trace 0.29038909 0.19 3 2 0.8932
Roy's Greatest Root 0.29038909 0.19 3 2 0.8932

No differences between times.

No differences in change over time between the groups
(compare to G-G time*group p-value=.6954)
 From rMANOVA analysis…
 Overall, are there significant differences between time
points?
 No, Time not statistically significant (p=.3287)
 Overall, are there significant changes from baseline?
 No, Time not statistically significant
 Do the two groups differ at any time points?
 Can’t say (never looked at raw scores, only difference values)
 Do the two groups differ in their responses over time?
**
 No, not even close; Group*Time (p-value=.89)
Can also test for the shape of
the response profile…

proc glm data=broad;

class group;
model time1-time4= group;
repeated time 3 polynomial /summary ;
run; quit;
 The GLM Procedure
Repeated Measures Analysis of Variance
Analysis of Variance of Contrast Variables

time_N represents the nth degree polynomial contrast for time

Contrast Variable: time_1

linear
Source DF Type III SS Mean Square F Value Pr > F

Mean 1 10.2083333 10.2083333 0.21 0.6716

group 1 21.6750000 21.6750000 0.44 0.5421
Error 4 195.7666667 48.9416667

Contrast Variable: time_2 quadratic

Source DF Type III SS Mean Square F Value Pr > F

Mean 1 84.37500000 84.37500000 3.80 0.1231

group 1 5.04166667 5.04166667 0.23 0.6586
Error 4 88.83333333 22.20833333

Contrast Variable: time_3

cubic
Source DF Type III SS Mean Square F Value Pr > F

Mean 1 130.2083333 130.2083333 5.88 0.0724

group 1 0.0750000 0.0750000 0.00 0.9564
Error 4 88.5666667 22.141666
 Can also get successive
paired t-tests
proc glm data=broad;
class group;
model time1-time4= group;
repeated time profile /summary ;
run; quit;

**Not adjusted for multiple comparisons!

 Repeated Measures Analysis of Variance
Analysis of Variance of Contrast Variables

time_N represents the nth successive difference in time

Contrast Variable: time_1 Time1 vs. time2

Source DF Type III SS Mean Square F Value Pr > F

Mean 1 6.00000000 6.00000000 0.35 0.5879

group 1 16.66666667 16.66666667 0.96 0.3823
Error 4 69.33333333 17.33333333

Contrast Variable: time_2 Time2 vs. time3

Source DF Type III SS Mean Square F Value Pr > F

Mean 1 192.6666667 192.6666667 2.56 0.1850

group 1 6.0000000 6.0000000 0.08 0.7918
Error 4 301.3333333 75.3333333

Contrast Variable: time_3 Time3 vs.

time4
Source DF Type III SS Mean Square F Value Pr > F

Mean 1 433.5000000 433.5000000 9.06 0.0395

group 1 0.1666667 0.1666667 0.00 0.9558
Error 4 191.3333333 47.8333333
Univariate vs. multivariate
 If compound symmetry assumption
is met, univariate approach has
more power (more degrees of
freedom).
 But, if compound symmetry is not
met, then type I error is increased
Summary: rANOVA and
rMANOVA
 Require imputation of missing data
 rANOVA requires compound
symmetry (though there are
corrections for this)
 Require subjects measured at same
time points
 But, easy to implement and
interpret
Practice: rANOVA and
rMANOVA

What effects do effects,

Within-subjects you
but no between-
expect to be
subjects effects.
statistically significant?
Time is significant.
Time?
Group?
Group*time is
significant.
Time*group?
Group is not
significant.
Practice: rANOVA and
rMANOVA

Between group
effects; no within
subject effects:
Time is not
significant.
Group*time is not
significant.
Group IS
significant.
Practice: rANOVA and
rMANOVA
Some within-
group effects, no
between-group
effect.
Time is
significant.
Group is not
significant.
Time*group is not
significant.
References
 Jos W. R. Twisk. Applied Longitudinal Data Analysis for Epidemiology: A
Practical Guide. Cambridge University Press, 2003.