Timeseries v' Prepost Design: Statistical Issues

1
(No Transcript)
2
Time-series v. Pre-post DesignStatistical Issues

• Analytic strategy (crude, refined)
• Correlation
• Within community (ICC), pair (r)
• Detectable effects (crude)
• Common elements, key differences
• Analytic options
• Trade-offs
• Validity, generalizability, power

3
Crude Analysis Pre-post (3 v 3)
MVPA(min/wk)
Baseline (n/2) 1 yr (n/4) 2 yr (n/4)
Community endpoint ? Change in mean MVPA H0
0 Mean D (Intervention) Mean D
(Control) Analysis Two-sample t-test (4 df) OR
... Paired t-test (2 df)
4
Crude analysis Time-series (6 v 0)
MVPA(min/wk)
2 yr (n/2) 2 yr (n/2)
Community endpoint ? Change in mean MVPA H0
0 Mean D Analysis One-sample t-test (5 df)
5
Fancier Analysis Time series (6 v 0)
MVPA(min/wk)
2 yr (n/2) 2 yr (n/2)
Community endpoint Trend in mean MVPA H0
Trend unchanged Analysis ...
6
Detectable Effect CalculationCommon Elements

• 80 power
• Crude analysis
• H0 tested with critical p-value 0.05
• Standard deviation 101 min/wk MVPA
• Community clustering of MVPA
• ICC (intraclass correlation)
• ____Community variance____ 0.009
• Community person variance
• Correlation between community pairs r ?

7
Detectable Effects Do the Math
Detectable diff in Pre-post Pre-post Time Mean
? paired unpaired series SD
(min/wk) 101 101 101 (ta/2,df tb,df) 2 df 4
df 5 df 4 (1ICC) 0.009 0.009 0.009
groups compared 2 2 1 communities/group 3 3 6
subjects/community n n n (1 r) ½ ? 0 0
8
Detectable Effects Bottom Line
Design Analysis Min/wk 6n3000 6n6000 Ti
me series One-sample t 12.8 9.1 Pre-post Indep-sa
mple t 27.3 19.3 Pre-post Paired t, r
0 39.4 27.8 , r
0.2 35.2 24.9 , r
0.5 27.8 19.7
9
Trade-offs All Else Equal
Design feature Guards Time Pre- against series
post Random assignment Confounding X X Repeated
measures Variability Community
pairing X ? Controls Secular trend X Detectabl
e effect Futility X Analytic refinement X