There is no quality control embedded in the program (as in the case of the Excel template). However, the R2 value has typically been above 95% for most datasets; when lower, it has been due to variation in the data and not a poor fit. HEPB also includes the residuals from the regression in the output. The speed of the program was determined by running it on a dataset with 5000 pairs of values (dataset XII, Table 1) on a Dell Optiplex 980 computer with Intel Core™ i7 CPU 860 @ 2.80 GHz processor, 8.00 GB of RAM, running on 64-bit, Microsoft Windows 7 Professional operating system, and the analysis was completed in 58 s. On a less powerful machine (Intel Core2

Duo E7500 @2.93GHz, 4 GB RAM, 32 bit Windows Selleckchem Z VAD FMK 7), it took 3 min and 56 s. When the estimation involves a single value, it is customary to construct a confidence interval around

the point estimate. This requires knowledge of the distribution that the estimate is expected to follow, and the width of a given confidence interval depends on the level of assurance required in ensuring that the unknown true value of the estimate resides within that interval. When the confidence interval is constructed for www.selleckchem.com/products/Romidepsin-FK228.html each Ŷ value in a regression, however, the two series of values at each end of the confidence interval then lie on either side of the Ŷ values (the regression line), thus forming a band along the length of the regression line. When the goal is to predict a new individual value of Y for a given value of X, sP(Ŷ), the standard error of Ŷ, is given as the square-root of the following expression ( Snedecor & Cochran, 1980): equation(2) sP2Y^=1n−2∑iny2−∑inxy2∑inx21+1n+x2∑1nx2;yi=Yi−Y¯,xi=Xi−X¯. The lower and upper prediction band limits for a given Ŷ value are obtained using Thalidomide the following equation: equation(3) Y^±tα,n−2sPY^where α is the level of significance and n is the sample size in terms of the number of

pairs of values. If the predictions are being made for k new X values, it would be necessary to use the Bonferroni inequality and obtain the t value from the Student’s t tables for α/k and (n − 2) degrees of freedom ( Snedecor & Cochran, 1980). However, since the purpose of drawing the prediction band in the present case is to give cut-off values that allow us to distinguish among sensitive, normal and resistant responses to a given anesthetic being used in any given experiment for the X values already in the data ( Fig. 3), Eq.  (5) is used to obtain the lower and upper limits of the prediction band. The c and d values for the upper and lower limits of the prediction band are estimated in the same manner of sequential sets of iterations as in the estimation of these parameters for the main regression equation, with the exception that the values of the corresponding prediction limits are used here instead of the observed values of the response variable.