Advanced Biostatistics - Simplified
Download as PPTX, PDF5 likes3,230 views
The document presents an overview of biostatistics, including fundamental concepts such as hypothesis testing, statistical significance, and data analysis methods. It discusses the importance of understanding different statistical tests and how to choose the appropriate one based on the type of data and research questions. Additionally, it emphasizes the need for proper sample size, distribution characteristics, and consideration of bias in biostatistical analysis.
Advanced Biostatistics - Simplified
- 1. 1 Biostatistics Simplified PREPARED & PRESENTED BY: DR. M. ALHEFZI DR. N. ALOTAIBI DR. A. KHALAWI DR. B. ALHEJAILI DR. M. ALGOTHAMI DR. S. ALGHAMDI SBCM | R1 | Taif A d v a n c e d
- 2. 2 SBCM | R1 | Taif Summarization Analysis – inference. WHY BIOSTAT ?! Collection Interpretation of the results Abhaya Indrayan (2012). Medical Biostatistics. CRC Press. ISBN 978-1-4398-8414-0. (QR-code above).
- 3. 3 Philosophy behind Hypothesis What is a hypothesis? CHANCE?! Mill’s Cannons / Methods – Agreement, Difference, Concomitant, Residues SBCM | R1 | Taif
- 4. 4 Am I right or wrong ?! Is it the truth ?! SBCM | R1 | Taif
- 5. 5 • BIAS? CONFOUNDING? CHANCE? CAUSE / EFFECT? • GENERALIZABILITY! • • SIGNIFICANCE SBCM | R1 | Taif •
- 6. 6 My Hypothesis Ha TEST! SBCM | R1 | Taif
- 7. 7 SBCM | R1 | Taif
- 8. 8 In other words … SBCM | R1 | Taif
- 9. 9 So, what language do we speak in biostat? SBCM | R1 | Taif MATH? MEAN, MEDIAN, MODE, RANGE … AREA UNDER THE CURVE, VARIANCE, SD … MEDICINE? EXPOSURE, DISEASE, OUTCOME, EFFECTIVITY, PREVENTION RELATIVE RISK, ABSOLUTE RISK
- 10. 10 MEDIAN. Biostatisticians’ language MEAN (μ). MODE. AREA UNDER THE CURVE: SBCM | R1 | Taif Variance. SD (σ).
- 11. 11 Biostatisticians’ language Standard Deviation (SD) SBCM | R1 | Taif
- 12. 12 SBCM | R1 | Taif Photo courtesy of Judy Davidson, DNP, RN
- 13. “ 13 WE MAKE MISTAKES! ” IN ORDER TO AVOID THEM, WE NEED TO SET RANGES FOR CHANCE, ALSO SET OUR CRITICAL LIMITS. TO END UP WITH A MASTERPIECE OF EVIDENCE! p-value SBCM | R1 | Taif H0 CI * vs. α level
- 14. 14 SBCM | R1 | Taif
- 15. 15 Test Hypothesis SBCM | R1 | Taif
- 16. 16 SBCM | R1 | Taif STEPS. Test Hypothesis ASSUMPTIONS. TESTS.
- 17. 17 LARGE SAMPLE SIZE. NORMAL DISTRIBUTION. Gaussian Dist. NO MULTICOLINIARITY. KNOWN & σ ). ASSUMPTIONS HOMOGENEITY. Test Hypothesis INDEPENDENCY. – Differs for each test. SBCM | R1 | Taif (μ
- 18. 1) RQ ? 2) H0 & H 1 3) TEST & ASSUMPTIONS. Test Hypothesis 4) α LEVEL, P-VALUE. – 7 steps of hypothesis testing. 5) TEST STATISTIC (DF). 6) DECISION. 7) CONCLUSION (YES/NO). STEPS SBCM | R1 | Taif 18
- 19. 19 Test Hypothesis TEST STATISTICS SBCM | R1 | Taif
- 20. 20 Each member in this group is exclusively linked to it Dependency Concept SBCM | R1 | Taif Output changes whenever input do so
- 21. Data Analysis • • • • SBCM | R1 | Taif Randomization. Restriction. Matching. Stratification. 21
- 22. 22 Statistical Tests SBCM | R1 | Taif
- 23. 23 Statistical Tests SBCM | R1 | Taif
- 24. 24 Choosing a Bivariate test Dependent VA (outcome, output) Indep. VA Input exposure 2 Cat. SBCM | R1 | Taif >2 Cat. Continuous Cat. χ2 χ2 t-test > 2 Cat. χ2 χ2 ANOVA Continuous t-test ANOVA Correlation Linear Regression
- 25. 25 Continuous Data SBCM | R1 | Taif
- 26. 26 Ordinal Data SBCM | R1 | Taif
- 27. 27 Categorical Data SBCM | R1 | Taif
- 28. Choosing the Best Statistical Test 28 Comparison the difference between groups Cat. VA (2) Cont. VA Independent sample (t-test) Mann-Whitney (U test) Cont. Dep. VA same group Paired Sample (t-test) Wilcoxon Cat. VA (>3) Cont. VA One Way ANOVA Kruskal Wallis Cat. VA Cat. VA Chi-Square (χ2 ) McNemar Association / Strength of Relationship Cont. VA Cont. VA Pearson (r) SBCM | R1 | Taif Spearman’s ρ Prediction Cont. VA Cont. or Cat. MLR SLR (Bivariate) PMT Cont. VA Cont. + Other VAs NPMT Cat. VA >1 Other VAs Logistic Regression By @alhefzi
- 29. 29 SBCM | R1 | Taif
- 30. 30 SBCM | R1 | Taif
- 31. 31 Considerations Normal Distribution & Sample Size. Large sample size (). Otherwise, do (Kolmogorov Smirnov) to check normality. Shape by inspection. If NPMT with Large sample size () less powerful than a PMT. Gaussian Distribution (). PMT with Non-Gaussian distribution () CLT. SBCM | R1 | Taif NPMT with Gaussian distribution, “small” sample size (). (small, Non-Gaussian) ( p-value). PMT with Non-Gaussian distribution, “small” sample size () CLT won’t work, inaccurate p-value.
- 32. 32 Considerations 1 or 2 sided p-value H0 (). Question: WHICH p-value is larger and why? (1 or 2 sided)? Based on: equal population means. Otherwise, any discrepancy is due to chance!! i.e. when formulating your Ha; consider “larger” critical p-value accordingly! Go for 1 sided (if) You have formulate a “directional” hypothesis. Set it BEFORE data collection. Otherwise, you will have to attribute the difference to chance. Go for 2 sided (if) SBCM | R1 | Taif Unsure or in doubt of your hypothesis direction. Set it BEFORE data collection. Otherwise, you will have to attribute the difference to chance.
- 33. The critical value is the number that separates the “blue zone” from the middle (± 1.96 this example). In a t-test, in order to be statistically significant the t score needs to be in the “dark-blue zone”. If α = .05, then 2.5% of the area is in each tail 2-tailed test Biostatisticians’ language SBCM | R1 | Taif 33
- 34. The critical value is either + or -, but not both. e.g. in a t-test In this case, you would have statistical significance (p < .05) if t ≥ 1.645. 1-tailed test Biostatisticians’ language SBCM | R1 | Taif 34
- 35. 35 Any number squared is a positive number. Therefore, area under the curve starts at 0 and goes to infinity (∞). To be statistically significant, needs to be in the upper 5% (α = .05). Compares observed frequency to what we expected. Chi-Square (χ2) – as an example Biostatisticians’ language SBCM | R1 | Taif Published on STAT 100 - Statistical Concepts and Reasoning (QR-code above)
- 36. 36 Considerations Regression or Correlation Correlation Regression X&Y are important to be set Swapping X&Y in the curve gives different results In Gaussian distribution Pearson SLR, MLR NPMT Spearman’s rho Logistic Regression Cause-effect relationship SBCM | R1 | Taif
- 37. @alhefzi End of Part I Thank you… QUESTIONS? 37