Slide 1 Delaware Chapter ASA January 19, 2006 Tonight’s speaker: Dr. Bruce H. Stanley DuPont Crop Protection “Applications of Binomial “n” Estimation, Especially when No Successes Are Observed” Del. Chapter of the ASA B. H. Stanley, 19 Jan 2006 Slide 2 Applications of Binomial “n” Estimation, Especially when No Successes Are Observed Dr. Bruce H. Stanley DuPont Crop Protection Stine-Haskell Research Center Newark, Delaware Tel: (302)-366-5910 Email: Bruce.H.Stanley-1@usa.dupont.com Del. Chapter of the ASA B. H. Stanley, 19 Jan 2006 Slide 3 Applications of Binomial “n” Estimation, Especially when No Successes Are Observed - Dr. Bruce H. Stanley – Many processes, such as flipping a coin, follow a binomial process where there is one of two outcomes. The researcher often knows that both outcomes are possible, even if no events of one of the outcomes is observed. This talk presents techniques for estimating number of trials, e.g., number of flips, based upon the observed outcomes only, and focuses on the case where events of only one possibility are observed. Dr. Stanley then discusses applications of this methodology. Del. Chapter of the ASA B. H. Stanley, 19 Jan 2006 Slide 4 Agenda • Introduction • Binomial processes • Replicated observations • Successes in at least one replicate • All replicates had no successes • All replicates are the same • Over and under dispersion • Some applications • Conclusion Del. Chapter of the ASA B. H. Stanley, 19 Jan 2006 Slide 5 Example: Codling Moth (Cydia pomonella (L.)) in Apples From: New York State Integrated Pest Management Fact Sheet http://www.nysipm.cornell.edu/factsheets/treefruit/pests/cm/codmoth.html Del. Chapter of the ASA B. H. Stanley, 19 Jan 2006 Slide 6 Typical Questions How many apples? How many “bad” apples? Del. Chapter of the ASA B. H. Stanley, 19 Jan 2006 Slide 7 Binomial Moments Let: Xi X s Number of successes for replicate I Average of Xis (i=1 to m) Sample standard deviation of Xis Mean X n p Variance Del. Chapter of the ASA s 2 n p (1 p) B. H. Stanley, 19 Jan 2006 Slide 8 Method of Moments Estimator (MME) Binomial Parameter n Xi X s n Estimator Let: Number of successes for replicate i Average of Xis (i=1 to m) Sample standard deviation of Xis 2 X n 2 X s Conditions 1. X > 0 2. X >s2 3. n> Xmax Del. Chapter of the ASA Note: >2, since 2 = np(1-p) = (1-p) B. H. Stanley, 19 Jan 2006 Slide 9 What About Over-dispersion? Del. Chapter of the ASA B. H. Stanley, 19 Jan 2006 Slide 10 Genesis “A simple model, leading to the negative binomial distribution, is that representing the number of trials necessary to obtain m occurrences of an event which has constant probability p of occurring at each trial.” (Johnson & Kotz 1969) Del. Chapter of the ASA B. H. Stanley, 19 Jan 2006 Slide 11 Negative Binomial Moments Let: Xi X s Number of successes for replicate i Average of Xis (i=1 to m) Sample standard deviation of Xis Mean X n p Variance Del. Chapter of the ASA s 2 n p (1 p) B. H. Stanley, 19 Jan 2006 Slide 12 Method of Moments Estimator (MME) Negative Binomial Parameter n Xi X s n Estimator Let: Number of successes for replicate i Average of Xis (i=1 to m) Sample standard deviation of Xis 2 X n 2 s X Conditions Del. Chapter of the ASA _ 1. X > 0_ 2. S2 > X 3. n > Xmax Note: 2> , since 2 = np(1+p) = (1+p) B. H. Stanley, 19 Jan 2006 Slide 13 Use the Var/Mean to Select a Method • If Mean > Variance use binomial • If Mean < Variance use negative binomial Del. Chapter of the ASA B. H. Stanley, 19 Jan 2006 Slide 14 What if…X =0 ? Del. Chapter of the ASA B. H. Stanley, 19 Jan 2006 Slide 15 Example Simulations Del. Chapter of the ASA B. H. Stanley, 19 Jan 2006 Slide 16 Example: Minitab – binomial variates (n=20, p=0.1) Del. Chapter of the ASA B. H. Stanley, 19 Jan 2006 Slide 17 Histograms of Generated Data Binomial Random Variates (n = 20, p = 0.1) 0 Bin(20,0.1) 10 reps 1 2 3 4 5 6 7 Bin(20,0.1) 20 reps Bin(20,0.1) 100 reps 30 20 Percent 10 Bin(20,0.1) 200 reps Bin(20,0.1) 1000 reps 30 0 1 2 3 4 5 6 7 0 20 10 0 0 Del. Chapter of the ASA 1 2 3 4 5 6 7 B. H. Stanley, 19 Jan 2006 Slide 18 Summary of Generated Data n=20,p=0.1 Reps Mean Variance 10 1.7 2.0111 20 2.05 1.5237 100 2.05 1.9268 200 1.95 1.8266 1000 2.013 1.6785 Theoretical 2 1.8 Del. Chapter of the ASA Est(n) -9.289617 7.98499 34.1112 30.81442 12.11411 NB Est(n) 9.289617 -7.98499 -34.1112 -30.81442 -12.11411 B. H. Stanley, 19 Jan 2006 Slide 19 Example: Minitab – binomial variates (n=1000, p=0.1) Del. Chapter of the ASA B. H. Stanley, 19 Jan 2006 Slide 20 Histograms of Generated Data Binomial Random Variates (n = 1000, p = 0.1) 72 n=10 80 88 96 104 112 120 128 n=20 n=100 20 15 10 Percent 5 n=200 20 n=1000 72 80 88 96 104 112 120 128 0 15 10 5 0 72 Del. Chapter of the ASA 80 88 96 104 112 120 128 B. H. Stanley, 19 Jan 2006 Slide 21 Summary of Generated Data n=1000,p=0.1 Reps Mean Variance 10 102.5 161.17 20 98.5 57.421 100 102.22 87.3 200 100.05 82.15 1000 100.25 80.38 Theoretical 100 90 Del. Chapter of the ASA Est(n) -179.0736 236.1852 700.3303 559.218 505.7908 NB Est(n) 179.0736 -236.1852 -700.3303 -559.218 -505.7908 B. H. Stanley, 19 Jan 2006 Slide 22 Key References Binet, F. E. 1953. The fitting of the positive binomial distribution when both parameters are estimated from the sample. Annals of Eugenics 18: 117-119. Blumenthal, S. and R. C. Dahiya. 1981. Estimating the binomial parameter n. JASA 76: 903 – 909. Olkin, I., A. J. Petkau and J. V. Zidek. 1981. A comparison of n estimators for the binomial distribution. JASA 76: 637 – 642. Johnson, N. L. and S. Kotz. 1969. Discrete Distributions. J. Wiley & Sons, NY 328 pp. (ISBN 0-471-44360-3) Del. Chapter of the ASA B. H. Stanley, 19 Jan 2006 Slide 23 Conclusions • You can work backwards from binomial data to estimate the number of trials. • If data appear “over-dispersed”, try the negative binomial distribution approach. • Bias adjustments exist. • Methods exist to handle the case where no events are observed. However, one must assume something about the probability of an event. Del. Chapter of the ASA B. H. Stanley, 19 Jan 2006 Slide 24 Thank You! Dr. Bruce H. Stanley DuPont Crop Protection Stine-Haskell Research Center Newark, Delaware Tel: (302)-366-5910 Email: Bruce.H.Stanley-1@usa.dupont.com Del. Chapter of the ASA B. H. Stanley, 19 Jan 2006 Slide 25 Delaware Chapter ASA Next Meeting: Feb 16, 2006 Professor Joel Best Author of “LIES, DAMN LIES, AND STATISTICS” and “MORE LIES, DAMN LIES, AND STATISTICS” Del. Chapter of the ASA B. H. Stanley, 19 Jan 2006