|
Can someone help ? I am getting a disagreement between lme() in Splus and what [ I hope ! ] is the equivalent fit using JMP. Everything agrees except the degrees of freedom assigned to the standard error to be used in the t-test (and confidence interval) for the slope term. JMP says 3 df and lme() says 1 df. ( Differences in parameterisation lead to numerically different fixed-factor effects, but the fitted values are identical). Full details follow ... if anyone can point me in the right direction I'd be very grateful. I have the dataset called 'TF' :- Strength type AnalDay Assay StorageDays AssayOcc 1 4 FinalSol 0 40.4355 0 1 2 4 FinalSol 4 41.0857 4 2 3 4 FinalSol 7 40.8318 7 3 4 10 FinalSol 0 103.6812 0 1 5 10 FinalSol 4 104.5013 4 2 6 10 FinalSol 7 104.1406 7 3 7 20 FinalSol 0 195.6095 0 1 8 20 FinalSol 4 196.6439 4 2 9 20 FinalSol 7 196.1706 7 3 and I'm trying to fit the model Assay = Strength + AssayOcc + StorageDays where Strength is a fixed effect factor, StorageDays is a fixed slope effect per day , assumed common to all 3 strengths AssayOcc is random effect for variability between assay occasions SPLUS :- > lme(fixed = Assay ~ Strength + StorageDays, data = "" random = ~ 1 | AssayOcc) Linear mixed-effects model fit by REML Data: TF AIC BIC logLik 18.22203 15.87866 -3.111015 Random effects: Formula: ~ 1 | AssayOcc (Intercept) Residual StdDev: 0.4563897 0.09662428 Fixed effects: Assay ~ Strength + StorageDays Value Std.Error DF t-value p-value (Intercept) 113.4024 0.4309204 4 263.163 <.0001 Strength1 31.6617 0.0394467 4 802.645 <.0001 Strength2 41.2318 0.0227746 4 1810.431 <.0001 StorageDays 0.0751 0.0925766 1 0.811 0.5661 Standardized Within-Group Residuals: Min Q1 Med Q3 Max -1.019119 -0.107893 -0.0007243514 0.08046544 1.127012 Number of Observations: 9 Number of Groups: 3 > intervals.lme(TF.lme) Approximate 95% confidence intervals Fixed effects: lower est. upper (Intercept) 112.205987 113.40241396 114.598841 Strength1 31.552162 31.66168333 31.771205 Strength2 41.168540 41.23177222 41.295005 StorageDays -1.101194 0.07510225 1.251399 Random Effects: Level: AssayOcc lower est. upper sd((Intercept)) 0.1115825 0.4563897 1.866705 Within-group standard error: lower est. upper 0.04832077 0.09662428 0.193214 JMP (also REML) :- 
Note the 3 df in the Denominator for 'StorageDays' in the Effects Tests table, and the consequently narrower confidence interval in the Parameter Estimates table. thanks Bob 
|