Presentation on theme: "Biostatistics course Part 17 Non-parametric methods Dr. C. Nicolas Padilla Raygoza Department of Nursing and Obstetrics Division of Health Sciences and."— Presentation transcript:
Biostatistics course Part 17 Non-parametric methods Dr. C. Nicolas Padilla Raygoza Department of Nursing and Obstetrics Division of Health Sciences and Engineers Campus Celaya-Salvatierra University of Guanajuato
Biosketch Médico Cirujano por la Universidad Autónoma de Guadalajara. Pediatra por el Consejo Mexicano de Certificación en Pediatría. Diplomado en Epidemiología, Escuela de Higiene y Medicina Tropical de Londres, Universidad de Londres. Master en Ciencias con enfoque en Epidemiología, Atlantic International University. Doctorado en Ciencias con enfoque en Epidemiología, Atlantic International University. Profesor Asociado C, Department of Nursing and Obstetrics, Division of Health Sciences and Engineerings, Campus Celaya Salvatierra, University of Guanjuato. email@example.com
Competencies The reader will know the non-parametric methods and when he(she) can use them. He (she) will apply non-parametric methods in an appropriate form. He (she) can obtain a confidence interval in non-paramethric analysis He (she) will apply Wilcoxon sum rank test He (she) will apply Wilcoxon He (she) will apply r Spearman.
Introduction Parametric methods They are base in means, standard deviations or probabilities. The Normal distribution is not always appropriate To study variables with a few observations, Non-symmetrical distributions, or Variables that can have more than two values
Introduction (contd…) When this happens, we use other anaylisis methods Non-parametric methods They are not based in the same assumptions that parametric methods, but also have some assumptions.
Categories (ranking), means, medians Some non-parametric methods use rankings en lugar de los real values. Categories are use to compare data, more for their ranking that for their size. PatientGlucose in blood (mg/dl) 1135 2225 370 4100 5110 6150 790 8100 9170 1060 1180
Categories (ranking), means, median Ranked in ascending order PatientGlucose in blood (mg/dl)Ranking 10601 3702 11803 7904 41005 8 6 51107 11358 61509 917010 222511
Are mean and median equals? To use mean and confidence interval is adequate, the distribution of values should be symmetric. To the median and confidence intervals are adequate, no need for assumptions.
Using the order (ranking) instead of original values, reduces the need for assumptions about the distribution, the calculations are simpler and faster. The disadvantage is that the original values are lost. Thus, non-parametric methods are used only to test hypotheses, not for estimation purposes. Are mean and median equals?
Non-parametric methods SituationNon-paramethric method Paramethric methods One sampleWilcoxon signed rank test Z statistic ( t test) Two indpendent samples Wilcoxon sum rank test Z statistic for two independent samples (t test) Two paired samples Wilcoxon signed rank test Z-paired statistic (t-paired test) One sample, two quantitative variables Correlation coefficient of Spearrman Correlation coefficient of Pearson
Data of one sample The table show data of glucose levels in blood from 11 patients. We want to know if the mean is 100 mg/dl. PatientGlucose in blood (mg/dl)Ranking 10601 3702 11803 7904 41005 8 6 51107 11358 61509 917010 222511
Data of one sample Alternative no parametric test is Wilcoxon signed rank test. It can be used to evaluate if the values in the sample are centered in 100 mg/dl. This test does not require Normality of the distribution of data, but requires that the distribution is symmetrical, but not necessarily take the form of "bell" as Normal.
Data of one sample Wilcoxon signed rank test is calculate by six steps: 1. To calculate the difference between each observation and the interest value, 100 mg/dl. 2. You should exclude any difference = 0. 3. To classify and order (ranking) differences by magnitude, not taken into accoun the sign. 4. Sum the rankings of positive differences. 5. Sum the rankings of negative differences. 6. Select the more little sums and call it T.
Data of one sample PatientGlucose in blood (mg/dl) Differences with 100 mg/dl Rnking 1060-406 370-304 1180-203 790-102 41000 8 0 5110101 1135355 6150507 9170708 22251259
Two independent groups 30 teenagers with acute apendicitis, were distributed 15 to underwent traditional apendicectomia and 15 with laparoscopic apedicectomia. For both groups, we evaluate post-surgical pain. Post-surgical painTraditionalLaparoscopy None13 Slight57 Moderate54 Severe41 Total15
Two independent groups To compate post-surgical pain in both groups, we can use Wilcoxon rank sum test. We define the null hypothesis H o : the two distributions overlap. We define alternative hypothesis H i : the two distributions are not overlap.
Two independent groups Wilcoxon rank sum test has three steps: We order the values in both groups in ascendant order. To calculate T as the sum of rankings of more short sample or one of two if the sample size is equal. To compare T-value in the critical values of Wilcoxon rank sum test.
Two independent groups Post-surgical painTraditionalLaparosco py Rankings None11+ None33 Slight59+ Slight715 Moderate521+ Moderate425 Severe429+ Severe130 Total15
Two paired groups The table show hours of improvement given by two analgesics in 12 patients with rheumatoid arthritis. To test that the improvement is the same with both analgesics, we can use paired-t test or Wilcoxon signed ranking test. With both methods, we calculate the difference of improvement in hours for each patient. PatientA AnalgesicB Analgesic 13.5 23.65.7 32.62.9 42.62.4 57.39.9 63.43.3 714.916.7 86.66.0 92.33.8 102.04.0 116.89.1 128.526.9
Two paired groups With Wilcoxon signed rank test, it is no requirement the Normality, but the data should be symmetrical to both sides of the median. H o : difference in medians = 0 H i = difference in medians ≠ 0 PatientA AnalgesicB AnalgesicDifferenceRankings 13.5 0 23.65.7-2.18 32.62.9-0.33 18.104.22.168 57.39.9-2.610 22.214.171.124 714.916.7-1.86 86.66.00.65 92.33.8-1.54 102.04.0-2.05 116.89.1-2.37 128.526.9-18.411
Two paired groups We calculate the Wilcoxon signed rank test for differences, making the following: 1.- Count how many differences non-zero. 2.- Order the differences by their magnitude, without take into account the sign. 3.- Sum rankings of positive differences. 4.- Sum rankings of negative differences. 5.- Select the more shor of the two sums and call it T. (Sum of negative differences = 59, sum of positive differences = 7, T=7). 6.- Compare the T-value in the critical values tables for Wilcoxon signed rank test. T=7, p<0.05.
Spearman’s correlation of ranks Table and graphic show incidence of colon cancer and average of meat intake per capita in 13 countries. Countr y Incidence colon ca Mean of intake of meat 1101 289 3115 4125 52233 66737 77332 8488 93741 103112 112129 12173 1331
Spearman ranks correlation It is appropiate for monotonic relationships, non- lineal. It is calculate at the same time that r’s Pearson, only using the rankings. To calculate it, we need three steps: To order the values of first variable, To order the values of second variable, To apply the formulae of r’s Pearson, using the rankings instead of original values.
Spearman ranks correlation Countr y Incidence colon ca Mean of meat intake Ranking of cancer Ranking of meat intake 110131 28927 311545 412554 52233811 6673712 773321310 8488116 937411013 10311298 11212979 1217363 133112
Comparison of methods ExampleParametric methodNon-parametric method Glucose in blood t test for one sample p>0.05 Wilcoxon signed rank test, p>0.2 Intensity of surgical pain t test for two independent samples p<0.05 Wilcoxon sun rank test p<0.05 Improvement of pain t paired test p>0.1Wilcoxon signed rank test, p<0.05 Corrlation between colon cancer and meat intake R Pearson, r= 0.65R Spearman, r=0.74
Bibliografy 1.- Last JM. A dictionary of epidemiology. New York, 4ª ed. Oxford University Press, 2001:173. 2.- Kirkwood BR. Essentials of medical ststistics. Oxford, Blackwell Science, 1988: 1- 4. 3.- Altman DG. Practical statistics for medical research. Boca Ratón, Chapman & Hall/ CRC; 1991: 1-9.