1. Exam 1 Median: 74 Quartiles: 68, 84 Interquartile range: 16 Mean: 74.9 Standard deviation: 12.5 z = -1: 62.4 z = -1: 87.4 z = -1z = +1 Worst Question: Operational definition (33/84)

2. Introduction to Hypothesis Testing: The Binomial Test 9/30

3. ESP Your friend claims he can predict the future You flip a coin 5 times, and he’s right on 4 Is your friend psychic?

4. Two Hypotheses Hypothesis – A theory about how the world works – Proposed as an explanation for data – Posed as statement about population parameters Psychic – Some ability to predict future – Not perfect, but better than chance Luck – Random chance – Right half the time, wrong half the time Hypothesis testing – A method that uses inferential statistics to decide which of two hypotheses the data support

5. Likelihood – Probability distribution of a statistic, according to each hypothesis – If result is likely according to a hypothesis, we say data “support” or “are consistent with” the hypothesis Likelihood for f(correct) – Psychic: hard to say; how psychic? – Luck: can work out exactly; 50/50 chance each time HypothesesStatistics Likelihood Support PopulationSample Probability Inference

6. Likelihood According to Luck Flip 12345123451234512345 YYYYYYNYYYNYYYYNNYYY YYYYNYNYYNNYYYNNNYYN YYYNYYNYNYNYYNYNNYNY YYYNNYNYNNNYYNNNNYNN YYNYYYNNYYNYNYYNNNYY YYNYNYNNYNNYNYNNNNYN YYNNYYNNNYNYNNYNNNNY YYNNNYNNNNNYNNNNNNNN f(Y)likelihood 51/32 45/32 310/32 2 15/32 01/32

7. Binomial Distribution Binary data – A set of two-choice outcomes, e.g. yes/no, right/wrong Binomial variable – A statistic for binary samples – Frequency of “yes” / “right” / etc. Binomial distribution – Probability distribution for a binomial variable – Gives probability for each possible value, from 0 to n A family of distributions – Like Normal (need to specify mean and SD) – n: number of observations (sample size) – q: probability correct each time

8. Binomial Distribution n = 5, q =.5n = 20, q =.5 n = 20, q =.25 Formula (optional): Frequency

9. Binomial Test Hypothesis testing for binomial statistics Null hypothesis – Some fixed value for q, usually q =.5 – Nothing interesting going on; blind chance (no ESP) Alternative hypothesis – q equals something else – One outcome more likely than expected by chance (ESP) Goal: Decide which hypothesis the data support Strategy – Find likelihood distribution for f(Y) according to null hypothesis – Compare actual result to this distribution – If actual result is too extreme, reject null hypothesis and accept altenative hypothesis “Innocent until proven guilty” – Believe null hypothesis unless compelling evidence to rule it out – Only accept ESP if luck can’t explain the data

10. Testing for ESP Null hypothesis: Luck, q =.5 Alternative hypothesis: ESP, q >.5 Need to decide rules in advance – If too extreme, abandon luck and accept ESP – How unlikely before we will give up on Luck? Where to draw cutoff? – Critical value: value our statistic must exceed to reject null hypothesis (Luck) Likelihood according to Luck: Binomial(n = 20, q =.5) Frequency Not at all unexpected Luck still a viable explanation Nearly impossible by luck alone

11. Another Example: Treatment Evaluation Do people tend to get better with some treatment? – Less depression, higher WBC, better memory, etc. – Measure who improves and who worsens – Want more people better off than worse Sign test – Ignore magnitude of change; just direction – Same logic as other binomial tests Count number of patients improved Compare to probabilities according to chance

12. Structure of a Binomial Test Binary data – Each patient is Better or WorseEach coin prediction is Correct or Incorrect Population parameter q – Probability each patient will improveProbability each guess will be correct Null hypothesis, usually q =.5 – No effect of treatmentNo ESP – Better or worse equally likelyCorrect and incorrect equally likely Alternative hypothesis, here q >.5 – Effective treatment; more people improveESP; guess right more often than chance Work out likelihood according to null hypothesis – Probability distribution for f(improve)Probability distribution for f(correct) Compare actual result to these probabilities – If more improve than likely by chance,If more correct than likely by chance, accept treatment is usefulabandon luck and accept ESP Need to decide critical value – How many patients must improve?How many times correct? Frequency Probability

13. Errors Whatever the critical value, there will be errors – All values 0 to n are possible under null hypothesis – Even 20/20 happens once in 1,048,576 times Can only minimize how often errors occur Two kinds of errors: Type I error – Null hypothesis is true, but we reject it – Conclude a useless treatment is effective Type II error – Null hypothesis is false, but we don’t reject it – Don’t recognize when a treatment is effective

14. Frequency Null Hypothesis (Bogus Treatments) Critical Value and Error Rates Type I Errors Frequency ? ? ? ? ? ? ? ? ? ? ? ? ? Alternative Hypothesis (Effective Treatments) Type II Errors

15. Critical Value and Error Rates Frequency Null Hypothesis (Bogus Treatments) Type I Errors Frequency ? ? ? ? ? ? ? ? ? ? ? ? ? Alternative Hypothesis (Effective Treatments) Type II Errors Increasing critical value reduces Type I error rate but increases Type II error rate (and vice versa) So, how do we decide critical value? Two principles – Type I errors are more important to avoid – Can’t figure out Type II error rate anyway Strategy – Decide how many Type I errors are acceptable – Choose critical value accordingly

16. Controlling Type I Error Type I error rate – Proportion of times, when null hypothesis is true, that we mistakenly reject it – Fraction of bogus treatments that we conclude are effective Type I error rate equals total probability beyond the critical value, according to null hypothesis Strategy – Decide what Type I error rate we want to allow – Pick critical value accordingly Alpha level (  ) – Chosen Type I error rate – Usually.05 in Psychology – Determines critical value Frequency Null Hypothesis (Bogus Treatments) Type I Error Rate

17. Summary of Hypothesis Testing Determine Null and Alternative Hypotheses – Competing possibilities about a population parameter – Null is always precise; usually means “no effect” Find probability distribution of test statistic according to null hypothesis – Likelihood of the statistic under that hypothesis Choose acceptable rate of Type I errors (  ) Pick critical value of test statistic based on  – Under Null, probability of a result past critical value equals  Compare actual result to critical value – If more extreme, reject Null as unable to explain data – Otherwise, stick with Null because it’s an adequate explanation Frequency Null Hypothesis (Bogus Treatments) Type I Error Rate (  ) Reject Null Keep Null

18. Review To test whether infants can read, you show pairs of good and bad words, and count how many times the baby crawls to the good word. What’s the null hypothesis? A.Baby is more likely to crawl to good word B.Baby is more likely to crawl to bad word C.Baby is equally likely to crawl to either word D.Baby will not choose either word Teddy bear Monster

19. Review What would be a Type II Error for this experiment? A.Concluding babies can’t read when they really can B.Concluding babies can read when they really can’t C.Correctly concluding babies can read D.Correctly concluding babies can’t read Teddy bear Monster

20. Review Each baby gets 6 trials. If any baby chooses the good word ≥5 times, you declare (s)he can read. Here are the probabilities of what will happen, according to the null hypothesis: What’s the Type I error rate? A.0.09 B.0.11 C.0.82 D.0.89 Number of good words 0 1 2 3 4 5 6 Probability.02.09.23.31.23.09.02