we find the P-value <strong>for</strong> the F-test. We can look in Table 6(b) with numerator DF as4 and denominator DF as 15 to see that the P-value is (much) lower than 0.01, sinceF = 61.2465 is much higher than 4.17 from the table. In R the P-value could be foundas 1-pf(61.2465,4,15) with the result: 4.1237e-09. Hence the correct answer is 1:At least one mean value is different from one of the others since the P-value is ≈0Question VIII.3 (19) The 99% confidence interval <strong>for</strong> µ A −µ B is? We use the approachfrom page 266:38.75 − 36.25 ± t 0.005√9.567( 1 4 + 1 4 )that is, using 15 degrees of freedom <strong>for</strong> the t:38.75 − 36.25 ± 2.947The correct answer then becomes 5: 2.5 ± 6.44Exercise IX√9.567( 1 4 + 1 4 )Two different brands of tablets with the same active compound are compared withrespect to their solubility. For each of the two brands 10 tablets were investigated. Foreach tablet, percent solubility is measured after the tablet have been kept in 1000 mlde-ionized water <strong>for</strong> a while. One measurement failed, so the following values <strong>for</strong> %solubility were found:Brand F 45 47 48 49 49 50 52 52 53 54Brand G 48 48 49 49 52 54 54 55 55Question IX.1 (20) What are the following five numbers <strong>for</strong> Brand G: 0%-percentile,lower quartile Q 1 , median, upper quartile Q 3 , and 100%-percentile? With n = 9 wecould use the procedure given in page 30 a number of times. However, since the medianof these 9 numbers is the middle number, that is 52, the only possible correct answeris 2: 48, 49, 52, 54, 55. (And we don’t have to think about how to define the 0% and100% percentiles - which simply are defined as the minimum and maximum values)Question IX.2 (21) A 99% confidence interval <strong>for</strong> the difference between the twomeans, which is not based on an assumption about normality of the data, is wanted.The following R lines are run:x=c(45,47,48,49,49,50,52,52,53,54)y=c(48,48,49,49,52,54,54,55,55)k = 10000xsamples = replicate(k, sample (x, replace = TRUE))ysamples = replicate(k, sample (y, replace = TRUE))mymeandifs = apply(xsamples, 2, mean)-apply(ysamples, 2, mean)myquantiles=quantile(mymeandifs, c(0.005,0.01,0.025,0.05,0.25,0.5,0.75,0.95,0.975,0.99,0.995))round(myquantiles,2)10
The result, which then is the rounded (the R-function round is in the last line appliedto round to two decimal points) percentiles of the bootstrap distribution of differencesof means:0.5% 1% 2.5% 5% 25% 50% 75% 95% 97.5% 99% 99.5%-4.93 -4.64 -4.14 -3.79 -2.54 -1.67 -0.80 0.40 0.83 1.29 1.52The 99% confidence interval <strong>for</strong> the difference between the two means based on thisis? We use Section 9.5.2 in the R-note - telling us that we need to read off the 0.5%and 99.5% percentiles of the bootstrapped difference values. Hence the correct answeris 4: [−4.93, 1.52] (Note that if these computations are re-done in R, the results willnot be exactly the same due to simulation uncertainty)Question IX.3 (22) We want to test the following hypothesis on level α = 0.05 basedon the in<strong>for</strong>mation in the previous question:H 0 : µ F − µ G = −4H 1 : µ F − µ G < −4What is the conclusion and the most correct argument? We use Section 9.6.4 in theR-note - doing bootstrap based hypothesis testing <strong>for</strong> the 2-sample situation. We cansee that less then 5% of the bootstrap difference values are below −4 (since the 5%percentile is −3.79. Hence answer 3 is the correct one: The hypothesis is rejected sincethe P-value is less than 5%Exercise XIt is known that the quotient between surface ares and volume has an influence on howfast a tablet is dissolved. To investigate this the following data was collected:%dissolved(y)Surface area/Volume(mm 2 /mm 3 ) (x)0.60 0.75 0.90 1.05 1.5033.00 36.25 38.75 44.50 63.50In R the following has been run:x=c(0.6,.75,0.9,1.05,1.5)y=c(33,36.25,38.75,44.50,63.5)summary(lm(y~x))with the following result:Call:11