DOC PREVIEW
UNC-Chapel Hill STOR 151 - Experiments and Observational Studies

This preview shows page 1-2-19-20 out of 20 pages.

Save
View full document
View full document
Premium Document
Do you want full access? Go Premium and unlock all 20 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 20 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 20 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 20 pages.
Access to all documents
Download any document
Ad free experience
Premium Document
Do you want full access? Go Premium and unlock all 20 pages.
Access to all documents
Download any document
Ad free experience

Unformatted text preview:

Experiments and Observational StudiesRecall basic definitions:• Population• Sample• Response variable• Explanatory variable1234Articles in the New York Times —http://www.nytimes.com/2009/01/27/health/research/27long.html?scp=1&sq=pope%20air%20pollution&st=csehttp://www.nytimes.com/2008/11/02/us/02smog.html?scp=1&sq=donora&st=cse5Example 1. A clinical trial is set up to compare a proposed newdrug with a placebo (some inert substance that is given to makepeople think they are talking a drug when they are not). 50patients are recruited, and we randomly select 25 of them toreceive the new drug, the rest receiving the placebo. After someperiod of time the two groups are compared to see whether thegroup receiving the new drug has recovered from the diseaseunder study.Example 2. In an attempt to study the health effects of air pollu-tion, a group of researchs selected 6 cities in very different envi-ronments — some from an urban setting (e.g. greater Boston),some from a heavy industrial setting (e.g. eastern Ohio), somefrom a rural setting (e.g. Wisconsin). Altogether they selected8000 subjects from the 6 cities, and followed their health for thenext 20 years. At this time their health prognoses were comparedwith measurements of air pollution in the 6 cities.6Note: Example 2 is based on a real study which was called,believe it or not, the six cities study. The researchers were fromthe Harvard School of Public Health.Which of these is experimental, and which is observational?7Definitions:In an experiment, the researchers assign “treatments” to agroup of subjects, usually via some randomization mechanism.Here treatments may refer to treatments in the medical sense(e.g. different drugs), or different treatments in an agriculturalexperiment (e.g. different fertilizers), or different methods inan industrial experiment (e.g. whether to include a catalyst inan industrial chemical process), or many other things. The keypoint is that this is something deliberate that is assigned by theresearchers controlling the study, rather than left to chance.In an observational study, the researchers simply “observe”a group of subjects without actually “doing” anything to thesubjects.8Another example:A study in California showed that students who study a musicalinstrument have higher GPAs than students who do not, 3.59to 2.91. Of the music students, 16% had all As, comparedwith only 5% among the students who did not study a musicalinstrument.Does this prove cause and effect?9It was an observational study, and also a retrospective study,relying on past observations of students’ choices and grades.Such a study cannot prove that studying music improved thestudents’ grades. Nor can it prove that students with goodgrades are more likely to choose to study music (though thatseems a more plausible hypothesis).An experiment would assign students at random to one of twocurricula, one of which included music lessons, the other withoutmusic. Then, after a period of time, it would compare the GPAsof the two groups. If this kind of prospective study showed astatistically significant difference between the two groups, wewould probably be justified in concluding cause and effect.10Clinical trials are experiments, usually performed under tightlycontrolled conditions, to remove any possibility of bias.We prefer an experiment because there is much less danger oflurking variables, and it is therefore better for establishing causeand effect.However, it is not always possible or ethical to conduct an ex-periment of every question of interest to society.11The six cities study, and others that have similar objectives (in-cluding the recent one in New England Journal of Medicine) havecertain common aspects:• Observational because participants not obliged to follow cer-tain regimes• It was prospective (good) because all the criteria includingselection of subjects were laid out at the start of the study• Other known health factors (diet, smoking etc.) were takeninto account• But there is still some danger of lurking variables , e.g. theremay have been other environmental or socio-economic vari-ables that were responsible for the effects.12SurveysA sample survey is based on selecting people from some popu-lation and asking them questions.There is a distinction between a survey and a census — thelatter is supposed to be a complete enumeration. The bestknown example of a census is the U.S. census that takes placeevery ten years. In recent decades this has been conducted by amail questionnaire, with follow up for those people who do notrespond the first time round. A controversy in connection withthe census is whether to use sampling to adjust the results.13Designing a SurveyStep 1: Identify the population.This may not be trivial. For example, consider the statement(which may or may not be true) 65% of the people in ChapelHill are Democrats. What is the population referred to here?Step 2: Identify the sampling frame.Usually a physical list, e.g. a telephone directory (but maybe noteveryone is in the directory....)Step 3: Decide the sampling design.This refers to the actual rule for choosing a sample.14The simplest example of a sampling design is the simple randomsample or SRS. This means that one we have chosen a samplesize n, all possible samples of size n are equally likely to bechosen.A good way to do that is through a random number table.15Example of Random SamplingRecently there have been many allegations related to steroidabuse in Baseball, as a result of which the players’ union hasfinally agreed to allow some testing. This is an example of howsuch a testing procedure might work.Let’s suppose that up to 10 players from each team are to beselected “at random” for drug testing. How would we selectsuch a sample for the St. Louis Cardinals?Table shows the current roster, grouped into as pitchers (P),catchers (C), infielders (I), outfielders (O). (The numbers in theleft hand column are mine.)1601 P Boggs 19 C LaRue02 P Carpenter 20 C Molina03 P Franklin 21 I Barden04 P Garcia 22 I Glaus05 P Hawksworth 23 I K. Greene06 P Kinner 24 I T. Greene07 P Lohse 25 I Hoffpauir08 P Manning 26 I Kennedy09 P McClellan 27 I Pujols10 P Miller 28 I Ryan11 P Motte 29 O Ankiel12 P Perez 30 O Barton13 P Pineiro 31 O Duncan14 P Ring 32 O Ludwick15 P Scherer 33 O Mather16 P Thompson 34 O Schumaker17 P Wainwright 35 O Stavinoha18 P Wellemeyer17Various possible strategies:• Have the tester sit in the clubhouse


View Full Document

UNC-Chapel Hill STOR 151 - Experiments and Observational Studies

Download Experiments and Observational Studies
Our administrator received your request to download this document. We will send you the file to your email shortly.
Loading Unlocking...
Login

Join to view Experiments and Observational Studies and access 3M+ class-specific study document.

or
We will never post anything without your permission.
Don't have an account?
Sign Up

Join to view Experiments and Observational Studies 2 2 and access 3M+ class-specific study document.

or

By creating an account you agree to our Privacy Policy and Terms Of Use

Already a member?