Page 152 - Statistics for Environmental Engineers
P. 152

L1592_frame_C17  Page 149  Tuesday, December 18, 2001  1:51 PM






                       Theory: The Paired t-Test Analysis
                       Define δ as the true mean of differences between random variables y 1  and y 2  that were observed as
                       matched pairs under identical experimental conditions. δ will be zero if the means of the populations
                       from which y 1  and y 2  are drawn are equal. The estimate of δ is the average of differences between n
                       paired observations:

                                                   d =  --------- =  1 ∑ ( y 1,i –  y 2,i )
                                                       ∑d i
                                                             ---
                                                        n    n
                                                         d
                       Because of measurement error, the value of   is not likely to be zero, although it will tend toward zero
                       if δ is zero.
                        The sample variance of the differences is:

                                                              (
                                                        s d =  ∑ d i –  d) 2
                                                         2
                                                            -------------------------
                                                              n 1
                                                                –
                       The standard error of the average difference   is:
                                                         d
                                                           s =  -------
                                                                s d
                                                            d
                                                                n
                       This is used to establish the 1 – α confidence limits for δ, which are d ±  . The correctness of
                                                                                s t n−1,α/2
                                                                                 d
                       this confidence interval depends on the data being independent and coming from distributions that are
                       approximately normal with the same variance.



                       Case Study Solution
                       The differences were calculated by subtracting the electrode measurements from the Winkler measure-
                       ments. The average of the paired differences is:
                                          – (  0.4) ++  – (  0.5) +  – (  1.0) +  … +  – (  0.4)
                                                0
                                     d =  ------------------------------------------------------------------------------------------------------ =  – 0.329 mg/L
                                                        14
                       and the variance of the paired differences is:

                                 – [  0.4 –  – (  0.329)] + [ 0 ( – 0.329)] +  … +  – [  0.4 –  – (  0.329)] 2
                                                            2
                                               2
                                                   –
                             2
                            s d =  ----------------------------------------------------------------------------------------------------------------------------------------------------------- =  0.244 mg/L) 2
                                                                                          (
                                                       14 1
                                                         –
                       giving s d  = 0.494 mg/L. The standard error of the average of the paired differences is:
                                                        ------------- =
                                                    s =  0.494  0.132 mg/L
                                                     d
                                                          14
                       The (1 – α)100% confidence interval is computed using the t distribution with ν = 13 degrees of freedom
                       at the α/2 probability point. For (1 − α) = 0.95, t 13,0.025  = 2.160, and the 95% confidence interval of the
                       true difference δ is:
                                                 ds t 13,0.025   <  δ  <  d +
                                                   –
                                                     d               s d t 13,0.025
                       © 2002 By CRC Press LLC
   147   148   149   150   151   152   153   154   155   156   157