Class Notes (834,631)
Psychology (3,518)
PSY201H1 (61)
Lecture 9

# Lecture 9 - The t Statistic (note: a & d = answers to last multiple choice qs)

4 Pages
51 Views

Department
Psychology
Course
PSY201H1
Professor
Alison Luby
Semester
Fall

Description
PSY201 Lecture 9; Nov. 17, 2011 ○ Rmbr s = SS/(n-1) ○  s = √(SS/df) Introduction to the t Statistic The t Statistic: An Alternative to z • sMused more in terms of sample variance, since • When hypothesis testing: provides accurate & unbiased estimate of the ○ Use a sample mean M to approximate the pop variance σ2 pop mean µ. ○ Standard error, σM, measures how well M approximates µ ○  Estimated sample error = =  σM= σ/√n ○ Rmbr must know sample mean before ○ Then compare M to µ by computing the computing sample variance, so there is a relevant z-score test statistic restriction on sample variability  z = (M - µ)/ M ○ Only n - 1 scores are independent & free to  Use to determine whether the obtained vary dif is greater than expected by chance ○  n - 1 = degrees of freedom (df) for sample by looking at unit normal table (for a variance normal distribution) • Like z-score, t statistic allows researchers to us• t statistic (like the z-score) forms a ratio: sample data to test hypotheses about an ○ Numerator: Obtained dif btwn M unknown µ hypothesized µ • Particular advantage of t statistic: ○ Denominator: Estimated standard error ○ Doesn’t require any knowledge of the pop (measures how much dif expected by σwhich we typically don’t know chance) ○ So can be used to test hypotheses about a completely unknown pop, ie:  Both µ & σ unknown • A large value for t (a large ratio) indicates  Only available info about pop comes obtained difference btwn data & hypothesis > from sample expected if treatment has no effect (ie. is just • All required for a hypothesis test with t is a sample & a reasonable hypothesis about µ error) ○ W large sample, df large, so estimation is • When we don’t haveσcan estimate it using the very good sample variability  So t statistic will be very similar to z- ○ Much like estimating µ using M score • Can use t statistic when determining whether ○ W small samples, df small, so t statistic treatment causes a change in µ provides relatively poor estimate of z • Sample is obtained from the pop  treatment is • Can think of the t statistic as "estimated zscore." administered to sample ○ As usual, if resulting sample M is significantly different from original µ, can conclude that ; treatment has a significant effect • Like before, hypothesis test attempts to decide • Just like we can make a distribution of z-scores, btwn: we can make a t distr ○ H o Is it reasonable that the discrepancy ○ Complete set of t values computed for every possible random sample for a specific btwn M & µ is simply due to sampling error and not result of treatment effect? sample size n, or specific degrees of freedom ○ H 1 Is the discrepancy btwn M & µ more than df expected by sampling error alone? • As df approaches infinity, t distr  ie. Is M significantly different from µ approximates normal distr st • How well it approximates normal distr • Critical 1 step for t statistic hypothesis test: ○ Calculate exactly how much dif btwn M & µ is depends on df reasonable to expect ○ ie. There’s a “family” of t distrs so there’s a ○ But since σ is unknown, impossible to dif sampling distr of t for each possible df • Larger n  larger n – 1  better the t distr compute standard error of M (σ =Mσ/√n) as approximates the normal distr done w z-scores ○  t statistic requires to use sample • W small values for df  t distr flatter & more data’s variance, s to compute spread out than a normal distr estimated standard error, s M •  s M s/√n ○ ie. obtained dif btwn data & hypothesis (ratio in numerator) is significantly larger than dif expected if no treatment effect (standard error in the denominator) • When M >> µ (compared to estimated standard error), obtain large value for t, so can reject H 0 • When mean dif small compared to s , faMl to reject H0 • Example 1 ○ z distr (red) & t-distr (green) (n-1 = 4 = df) ○ t peak lower than z, sides higher than z • Researcher wanted to know whether people who (more density away from mean) practice yoga have dif (includes greater and lesser) depression level from general pop; uses a depression inventory that has mean score of 25 for pop • 14 indivs who regularly practice yoga complete the inventory • Scores: 15, 19, 24, 27, 18, 22, 23, 25, 17, 16, 20, 21, 17, 18 • Do the yoga students show a level of depression significantly dif from general pop? • • n = 14; µ = 25; M = 20.14; α = 0.05 ○ z distr (red) & t-distr (purple) (n-1= 40 = df) • Hypotheses: • As you can see, both distrs are unimodal & ○ H o µyoga= 25 symmetrical about the mean (= 0) ○ H 1 µyoga≠ 25 • However, have a dif shaped distrs when • Since we don’t know pop variance, have to use t small n statistic, so need to find df • And t has larger and more variable 2 ○ df = n – 1 = 13 variance, s • Now, determine our critical t value from the ○ ie. estimated standard error varies btwn table samples ○ t = 2.16  s = √(s /n) critical M • Then, determine the test statistic & reject H of >  As n & df ↑, variability in t distr ↓, tcritical becomes more normal ○ SS = 5852 – 79524/14 = 171.71 • To evaluate t statistic from hypothesis test: ○ Must select an α level ○ Estimated sample error = s = M = ○ Then find the appropriate value of df ○ Determine the critical t statistic from the t 0.9713 distr table, Appendix B, page 729  SS/13 = 13.21  #s in table are values of t that separate  13.21/14 = 0.94 ○ tobtained(20.14-25)/0.9713 = -5.003 the tail from the main body of the distr  Please look at this table to see that as df
More Less

Related notes for PSY201H1
Me

OR

Join OneClass

Access over 10 million pages of study
documents for 1.3 million courses.

Join to view

OR

By registering, I agree to the Terms and Privacy Policies
Already have an account?
Just a few more details

So we can recommend you notes for your school.