class: center, middle, title-slide .upper-right[ <!-- --> ] # Class 25 <br/> Effect sizes and *p*-hacking .title-hline[ ## November 27, 2017 ] <!-- --- --> <!-- class: middle, center, inverse --> <!-- # Effect sizes --> --- class: middle, center, inverse # Statistical errors and *p*-hacking --- # Issues with statistics in modern science * Over-reliance of *p*-values when determining an experiment's worth * Data dredging/*p*-hacking * Lack of transparency regarding statistical analysis * Poor statistical practices among researchers * Lack of reports about experiments that fail to reject the null hypothesis * Ignoring or underemphasizing effect size --- # .font70[Example: Which political party is better for the economy?] * Start with a reasonable hypothesis: the economy is affected by whether or not Democrats or Republicans are in office -- * Collect data about different measures of economic performance and when different politicians were in office -- * Construct a basic model connecting the two -- * [FiveThirtyEight Applet](http://53eig.ht/HackingScience) (<http://53eig.ht/HackingScience>) <!-- --- --> <!-- class --> <!-- middle, center, inverse --> <!-- # Data-driven modeling --> <!-- --- --> <!-- # .font80[Relationships between two — or more — variables] --> <!-- --- --> <!-- # Examples of trends between variables --> <!-- --- --> <!-- # .font80[What are we hoping to accomplish with modeling?] --> <!-- --- --> <!-- # Machine learning --> <!-- --- --> <!-- class: middle, center, inverse --> <!-- # Linear regression --> <!-- --- --> <!-- # The simplest model --> <!-- --- --> <!-- # Correlation --> <!-- --- --> <!-- # Linear regression by gut intuition --> <!-- --- --> <!-- # Residuals --> <!-- --- --> <!-- # Criterion for linear regression --> <!-- --- --> <!-- # Anatomy of a regression line --> <!-- --- --> <!-- # Prediction --> <!-- --- --> <!-- # Extrapolation --> <!-- --- --> <!-- # .font70[Conditions for using least-squares linear regression] --> <!-- --- --> <!-- # *R²* measure --> <!-- --- --> <!-- class: middle, center, inverse --> <!-- # Statistical inference for linear regression --> <!-- --- --> <!-- # Slope *p*-values --> <!-- --- --> <!-- # Slope confidence intervals -->