• Home
  • University of Queensland
  • Analysis Of Scientific Data
  • Hypothesis Testing and Statistical Analysis

Hypothesis Testing and Statistical Analysis

For the purposes of this guide, we will be using the Data.csv file. data will be the name of the variable we use for the csv. response_variable and explanatory_variable are named for the Data.csv file. Sometimes I may use different variable names for specific tests. This is for convenience. HYPOTHESIS TESTING Binomial probabilities dbinom(x=6, size=10, p =. 3) · gives P(X = 6) for X ~ Binomial (10, .3). pbinom(x=6, size=10, p =. 3) · gives P(X ? 6). sum(dbinom(7:10, size = 10, prob = 0.3)) · gives P(X?7). Note, you don't have to put x=, size=, or p =. Simply putting numbers in works. z values and significance qnorm(0.975) · gives the z statistic corresponding to quantile 0.975 on the Normal distribution. pnorm(1.96) . gives the quantile corresponding to the z statistic 1.96 on the Normal distribution. Finding t values for quantiles pt( t-statistic, df = _ ) · Gives the quantile corresponding to the t-statistic, given the degrees of freedom. qt( quantile, df = _ ) · Gives the standard deviations needed for that confidence quantile. t test t.test (response_variable ~ explanatory_variable, data = data, alternative = 'less' or 'greater') · The explanatory variable must have only two choices. Any more and you should refer to ANOVA. . This is the Welch approximation t test. For pooled t test, use 'var.equal = TRUE'. . The default t test is two sided, for a one-tailed test, use either 'less' or 'greater' depending on the alternative hypothesis. One-way ANOVA / regression ANOVA summary(aov(response_variable ~ explanatory_variable, data = data)) · Explanatory variable can be anything with two or more distinct categories. · One-way ANOVA or regression ANOVA depending on whether the explanatory variable is categorical or quantitative. pf(25, df1 = 2, df2 = 12) . Finds the p-value for an f-ratio of 25, a groups degrees of freedom of 2, and a residuals degrees of freedom of 12. Two-way ANOVA summary(aov(response_variable ~ explanatory_variable*explanatory_variable2, data = data)) . Note the explanatory_variable2 in testing a two-way ANOVA. Linear regression summary(lm(response_variable ~ explanatory_variable, data=data)) . Gives more information about the linear model, including standard errors, residual standard error, t-values, and two-sided p-values. . We care about the standard error and two-sided p-value for the slope if we are working out a test for association. · Can be used for both [quantitative x quantitative] or [quantitative x categorical] (which will be translated to 0 and 1). summary(lm(response_variable ~ explanatory_variable * explanatory_variable2, data = data)) · Allows testing of multiple explanatory variables in a linear regression association test. . Includes adjustment for an interaction effect between the explanatory variables. summary(lm(response_variable ~ explanatory_variable + explanatory_variable2, data = data)) · Drops the interaction between explanatory variables, useful if there is indeed no interaction. . Simplifies testing of two explanatory variables and their association with the response. Confidence/prediction intervals in linear regression predict(lm(response_variable ~ explanatory_variable, data=data), newdata=data.frame(explanatory_variable = []), interval = 'confidence') · A confidence interval, used for all individuals in a population with explanatory_variable = predict(lm(response_variable ~ explanatory_variable,