Textbook Notes (368,531)
Canada (161,958)
Statistics (133)
STAB22H3 (130)
Ken Butler (11)
Chapter 13

STAB22 - Highly detailed Chapter 13 Notes, Summer 2012, Ken Butler

13 Pages
Unlock Document

Ken Butler

CHAPTER 13 – EXPERIMENTS AND OBSERVATIONAL STUDIES WHERE ARE WE GOING? – 4 principles of experimental design – 1. control – 2. randomize – 3. replicate – 4. block if appropriate – together these 4 principles depict the ideal experiment & how to understand results that come from it – much of knowledge that is known about social sciences, & natural sciences derive from carefully designed expts MAIN BODY Example (Music) – study wants to look at who gets better grade: students who play musical instruments or those who do not – found that, by comparing the performance of those who play and those who do not, the ones that play have higher overall GPA – 3.59 vs. 2.91 – 16% of music students had As, while only 5% for non-music OBSERVATIONAL STUDIES [1] Example (Music) – study attempted to show assoc. b/ween music edu. & grades – simply observed students in their envirnmt, recorded the choices they made (music or non-music) and then outcome (GPA) OBSERVATIONAL STUDIES - ex. is Example (Music) - no assigning of choices is done - merely observing RETROSPECTIVE STUDY - observational study that involves - identifying subj's from records - looking at subjs' past results - Example (Music) - collected data about students' past grades, and saw whether or not they played music [2] Why can we not conclude that music education CAUSES good grades? - for music education to cause good grades, this requires that there are no other diff's b/ween the two groups being looked at that can account for the diff's in the grades. - for this Ex., cannot conclude causation b/c whether or not they studied music was NOT the only diff. b/ween the 2 groups [3] Consider lurking variables - Example (Music) - var's that are we have not taken into acc't may have caused groups to perform diffntly - ex. work ethic - ex. parental support - ex. wealth - ex. level of intelligence (prior to music education) ^- these are ex's of var's that could have caused diff's b/ween the performances of the two groups [4] RETROSPECTIVE STUDIES - restrospective records can potentially have errors, b/c its based on historical data - often are used to discover var's related to rare outcomes (Ex. specific diesases) - first identify ppl w/ disease - then look into history & heritage to find things related to this diseased condition - have restricted view of world b/c typically are restricted to small part of popn - ex. identifying ppl w/ that specific disease; only minority may even have it in the first plac e [5] PROSPECTIVE STUDIES - observational study that involves - identifying subj's in advance - collecting data as events occur - observes subj's over time - record var's of interest, and see what is outcome as it occurs - Example - Music - select students who are going to join music lessons but have not do so yet - track their academic performance over several yrs - compare these music students' results to another group of subj's that we track which aren't taking music lessons [6] OBSERVATIONAL STUDIES - CANNOT POSSIBLY DETERMINE CAUSATION - regardless of if PROSPECTIVE, or RETROSPECTIVE - doesn't guarantee that impt, related var's to the outcome are the most impt, or even CORRECT - Example - Music - non-music vs. music might still differ in an IMPT way that was failed to be observed - could be this unobserved var. that caused good grades, instead of music per se RANDOMIZED, COMPARATIVE EXPERIMENTS [1] EXPERIMENT - Example - Music - take a group of students - randomly assign half to take music lessons, and the other half never - then, compare their grades several yrs after [2] EXPERIMENT - REQ. random assignment of subj's to treatments - Examples of questions that a designed experiment can justify - does music lessons cause good grades? - does taking Vitamin C decrease the chance of getting a cold? - does working with computers improves performance in STAB22? - is this drug a safe, health-conscious treatment for this disease? [3] EXPERIMENT - studies relationship b/ween 2/more var's - must identify at least 1: - FACTOR - name for EXPLANATORY VARIABLE in expts - RESPONSE VARIABLE How does experiment differ from other forms of investigation? - expter.. - actively & intentionally manipulates factors to control details of possible treatments - randomly assigns subj's to treatments - then observes response var. & compares responses for diff groups of subj's who were treated diffntly (ie. were under diff. treatments) - ex. compare treatment A subj's response to treatment B's subj's Example - Sleep - design expt to see if amount of sleep & exercise affects performance => factors = sleep, exercise => response var = performance [4] EXPERIMENT What do we call the individuals we experiment on? = EXPERIMENTAL UNITS What is the alternative term to the above used when working with humans? = PARTICIPANTS or SUBJECTS - more likely to attract them as prospective individuals to expt on if we call them by this instead of calling them EXPERIMENTAL UNITS [5] EXPERIMENT - LEVELS = specific val's for the factor that are chosen by expter - each individual is assigned a level Example - Sleep - participants are assigned to either 4, 6, or 8 hours => 3 lvls in total, and each participant is assigned one of them Levels (con.) - often times there are several factors at variety of lvls - ex. factor 1 = sleep time - lvls = 4, 6, or 8 hours - ex. factor 2 = treadmill time - lvls = treadmill, no treadmill - TREATMENT = combination of specific lvls from all factors that an experiment al unit (aka subj. or participant) receives Example - Sleep - 6 possible treatments - 4 hrs, treadmill - 6 hrs, treadmill - 8 hrs, treadmill - 4 hrs, no treadmill - 6 hrs, no treadmill - 8 hrs, no treadmill [1], 344 How to assign subj's to treatments? - to derive a fair conclusion from expt, must assign participants to treatments RANDOMLY [2] Why random? - to ensure that results of experiment are valid AN EXPERIMENT (GREEN BOX) - MANIPULATEs factor lvls to create treatments - RANDOMLY ASSIGNs subj's to treatments - COMPAREs responses of subj groups across treatments - ex. compare Treatment A (4 hrs, treadmill) vs Treatment B (6 hrs, treadmill) BLUE BOX - clinical trials = randomized health experiments - ex. randomly assigned postmenopausal women to take either - a) hormone replacement therapy - b) inactive pill => they are either treated to therapy or no therapy (lvls of factor) - concluded that hormone replacement w/ estrogen INCR'ed stroke risk THE FOUR PRINCIPLES OF EXPERIMENTAL DESIGN [1], 345 Design of Experiments (1935) - depicts methodology for designing expts - by Sir R.A. Fisher - his ideas are summarized in 4 principles - 1. Control - 2. Randomize - 3. Replicate - 4. Block 1. Control - control sources of variation other than factors being tested by making conditions similar as possible for ALL treatment groups - controlling unrelated sources of variation decr's variability in response => makes it easier to detect diff's among treatment groups [?] [2] - it is risky to make generalizations from experiment to other lvls of a CONTROLLED FACTOR (ex) - we test 2 diff. laundry detergents and carefully control water temp at 82 Centigrade - => control means that we keep it constant - by doing this, it reduces variation in our results that is DUE to the controlled var. (ie. water temp in this case) - cannot also justify that b/c of keeping at this temp, if you keep at any other temp (ex. colder temp's), it will also reduce variation in results due to this controlled var. [3] We control two things, but for diff. reasons and ways: Control: factors - How: by assigning usbj's to diff. factor lvls - Why: to see how does response var. comparing one factor lvl to another Control: other sources of variation - How: ex. keep them constant across all treatments - Why: prevent them from changing & impacting response var. 2. Randomize [4] Randomization - permits us to equalize effects of unknown or uncontrollable sources of variation - doesn't eliminate these sources - what it does do is spread them out across all treatment groups evenly - cannot use statistics to draw conclusions from experiment IF - experimental units were NOT assigned to treatments at random - what is benefit of assigning at ran
More Less

Related notes for STAB22H3

Log In


Join OneClass

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

Sign up

Join to view


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.

Reset Password

Please enter below the email address you registered with and we will send you a link to reset your password.

Add your courses

Get notes from the top students in your class.