Understanding Statistics and Experimental Design. How to Not Lie With Statistics (Statistics and Experiment Planning for the Uninitiated)
This book is suitable even for beginners - the authors did their best to convey the basics of statistics as simply as possible, with a minimum of formulas and maximum meaning. If you’ve ever heard magic words like a/b tests, stats, t-criteria, p-value and wanted to understand what that means, this is the book for you. If you know theorver, matstats, calculus, then it will be easier to follow the thoughts of the authors, but even without this book will allow you to start feeling crap, which is sometimes shown under the guise of a great experiment. In the era of data-driven, this book is 150+ pages are very good:) She is. free On the Springer website.
The book itself consists of 3 parts 12 heads I - Principles of statistics This part must be read for people who use statistics to make decisions. 1 Basics of probability theory The authors talk about probability, probability distribution, conditional probability and the concept of independent events. Further, the analysis option for a certain disease is considered and the terms sensitivity are further introduced. (sensivity)specificity (specificity)Frequency of false positive results (false positive rate) Frequency of false negative results (miss rate). Here’s how it all works together and why even doctors don’t always understand the statistics.) 2 Experiment Planning and Basics of Statistics: Signal Detection Theory The authors go from the signal detection theory and on the example of a yellow submarine and sonar show what the design of the experiment and decision-making look like. In fact, we have signal and noise sensibility, and we have a certain threshold for decision-making. In the end, the percentage of correct decisions mixes them together. This conclusion is further shown in the part about the t-criterion, where the effect size is mixed with the sample size:) 3 Main concept of statistics Here the authors talk about the canonical approach to statistics, discuss statistics for the sample average, and then show how you can compare sample averages using one-sided and two-sided t-criteria. Further, the authors show that in a standard test, p-value controls the error rate of the first type, or false positive, when we find an effect and it does not. Even more interesting is the discussion of the power calculation of the experiment, which indicates the probability of obtaining a significant result if the alternative hypothesis is correct. (That is, the averages in our aggregates differ.). The concept of power is needed in later chapters, especially in the third part of the book. In fact, the most juicy part of this chapter is the consequences at the end, which directly affect the design of experiments and talk about sample size, effect size, zero results and so on. 4 - Variations on t-criteria This is a chapter with an asterisk relative to the third chapter where the usual t-criterion was considered.
II - Multiple Hypothesis TestingH is a very interesting part about how difficult it is to design a proper experiment with multiple hypothesis testing. 5 The problem of multiple hypothesis testing 6 - Dispersion analysis (ANOVA) 7 Experiment planning: model fit, power and complex plans - a very important chapter for the practitioner-experimenter, where the authors share the correct method and show by example how you can shoot yourself both legs and not only the best. 8 - Correlations. Comparison of t-criteria, ANOVA, and standard history to look for correlations between relative variables (In ANOVA, we have independent nominal variables)
Meta-analysis and the Crisis of Science This part is interesting to read to scientists and those who like to read the articles of English scientists:) Here, the authors on their fingers show how in pursuit of results, the authors of the studies publish too good results to be true:) 9 Meta-analysis. 10 - Reproducibility 11 - Values of excess success 12 Proposed improvements and unresolved problems
#Math #Statistics #PopularScience #Science #ML #Data