Hypothesis Testing Outline The Null Hypothesis Type I and Type II Error Using Statistics to test the Null Hypothesis The Logic of Data Analysis Research Questions and Hypotheses Research question: Non-directional: No stated expectation about outcome Example: Do men and women differ in terms of conversational memory? Hypothesis: Statement of expected relationship Directionality of relationship Example: Women will have greater conversational memory than men Grounding Hypotheses in Theory Hypotheses have an underlying rationale: Logical reasoning behind the direction of the hypotheses (theoretical rationale – explanation) Why do we expect women to have better conversational memory? Theoretical rationale based on: 1. Past research 2. Existing theory 3. Logical reasoning The Null Hypothesis Null Hypothesis - the absence of a relationship E..g., There is no difference between men’s and women’s with regards to conversational memories Compare observed results to Null Hypothesis How different are the results from the null hypothesis? We do not propose a null hypothesis as research hypothesis - need very large sample size / power Used as point of contrast for testing Hypotheses testing When we test observed results against null: We can make two decisions: 1. Accept the null No significant relationship Observed results similar to the Null Hypothesis 2. Reject the null Significant relationship Observed results different from the Null Hypothesis Whichever decision, we risk making an error Type I and Type II Error 1. Type I Error Reality: No relationship Decision: Reject the null Believe your research hypothesis have received support when in fact you should have disconfirmed it Analogy: Find an innocent man guilty of a crime 2. Type II Error Reality: Relationship Decision: Accept the null Believe your research hypothesis has not received support when in fact you should have rejected the null. Analogy: Find a guilty man innocent of a crime Potential outcomes of testing Decision Accept Null Reject Null R E No A Relationship 1 2 3 4 L I T Y Relationship Potential outcomes of testing Decision Accept Null Reject Null R E No A Relationship Correct decision 2 L I T Y Relationship 3 4 Potential outcomes of testing Decision Accept Null Reject Null R E No A Relationship 1 2 L I T Y Relationship 3 Correct decision Potential outcomes of testing Decision Accept Null Reject Null R E No A Relationship 1 Type I Error L I T Y Relationship 3 4 Potential outcomes of testing Decision Accept Null Reject Null R E No A Relationship 1 2 L I T Y Relationship Type II Error 4 Potential outcomes of testing Decision Accept Null Reject Null R E No A Relationship Correct decision Type I Error L I T Y Relationship Type II Error Correct decision Function of Statistical Tests Statistical tests determine: Accept or Reject the Null Hypothesis Based on probability of making a Type I error Observed results compared to the results expected by the Null Hypotheses What is the probability of getting observed results if Null Hypothesis were true? If results would occur less than 5% of the time by simple chance then we reject the Null Hypothesis Start by setting level of risk of making a Type I Error How dangerous is it to make a Type I Error: What risk is acceptable?: 5%? 1%? .1%? Smaller percentages are more conservative in guarding against a Type I Error Level of acceptable risk is called “Significance level” : Usually the cutoff - <.05 Conventional Significance Levels .05 level (5% chance of Type I Error) .01 level (1% chance of Type I Error) .001 level (.1% chance of Type I Error) Rejecting the Null at the .05 level means: Taking a 5% risk of making a Type I Error Steps in Hypothesis Testing 1. State research hypothesis 2. State null hypothesis 3.Set significance level (e.g., .05 level) 4. Observe results 5. Statistics calculate probability of results if null hypothesis were true 6. If probability of observed results is less than significance level, then reject the null Guarding against Type I Error Significance level regulates Type I Error Conservative standards reduce Type I Error: .01 instead of .05, especially with large sample Reducing the probability of Type I Error: Increases the probability of Type II Error Sample size regulates Type II Error The larger the sample, the lower the probability of Type II Error occurring in conservative testing Statistical Power The power to detect significant relationships The larger the sample size, the more power The larger the sample size, the lower the probability of Type II Error Power = 1 – probability of Type II Error Statistical Analysis Statistical analysis: Examines observed data Calculates the probability that the results could occur by chance (I.e., if Null was true) Choice of statistical test depends on: Level of measurement of the variables in question: Nominal, Ordinal, Interval or Ratio Logic of data analysis Univariate analysis One variable at a time (descriptive) Bivariate analysis Two variables at a time (testing relationships) Multivariate analysis More than two variables at a time (testing relationships and controlling for other variables) Variables Dependent variable: What we are trying to predict E.g., Candidate preference Independent variables: What we are using as predictors E.g., Gender, Party affiliation Testing hypothesis for two nominal variables Variables Null hypothesis Procedure Gender Passing is not related to gender Pass/Fail Chi-square Testing hypothesis for one nominal and one ratio variable Variables Null hypothesis Procedure Gender Score is not related to gender Test score T-test Testing hypothesis for one nominal and one ratio variable Variable Null hypothesis Procedure Year in school Score is not related to year in ANOVA school Test score Can be used when nominal variable has more than two categories and can include more than one independent variable Testing hypothesis for two ratio variables Variable Null hypothesis Procedure Hours spent studying Score is not related to hours spent studying Test score Correlation Testing hypothesis for more than two ratio variables Variable Null hypothesis Procedure Hours spent studying Score is positively related to hours Classes spent studying and Multiple missed negatively related regression to classes missed Test score Commonality across all statistical analysis procedures Set the significance level: E.g., .05 level Means that we are willing to conclude that there is a relationship if: Chance of Type I error is less than 5% Statistical tests tell us whether: The observed relationship has less than a 5% chance of occurring by chance Summary of Statistical Procedures Variables Nominal IV, Nominal DV Procedure Chi-square Nominal IV, Ratio DV T-test Multiple Nominal IVs, Ratio DV Ratio IV, Ratio DV ANOVA Multiple Nominal IVs, Ratio DV with ratio covariates ANCOVA Multiple ratio Multiple Regression Pearson’s R