Week 5 notes Stat121 :)
Week 5 notes Stat121 :) STAT 121
Popular in Principles of Statistics
Popular in Statistics
STAT 201 003
verified elite notetaker
This 5 page Class Notes was uploaded by Amanda Berg on Saturday October 3, 2015. The Class Notes belongs to STAT 121 at Brigham Young University taught by Dr. Christopher Reese in Fall 2015. Since its upload, it has received 51 views. For similar materials see Principles of Statistics in Statistics at Brigham Young University.
Reviews for Week 5 notes Stat121 :)
Report this Material
What is Karma?
Karma is the currency of StudySoup.
Date Created: 10/03/15
Week 5 Notes 1 Sampling Observational studies a YOU CANNOT CONCLUDE CAUSATION from observational studies ONLY experiments b Observation of individuals in a sample subset of population of interest c Use sample to describe the population i Why use a sample instead of taking a census of the entire population 1 Practical 2 Cheap 3 Often more accurate 2 Terminology Population entire group of individuals of interest Sample individuals that are selected and measured Variable characteristic that is measured Parameter numerical fact about the variable in the population Statistic corresponding numerical fact in the sample i Take sample statistics and say things about the population parameters f We don39t know the parameter when we are sampling 3 For sampling to work a Explicitly describe population b Explicitly describe variable c Select representative sample 4 Ways to sample BADLY a Convenience sampling i Select individuals in the easiest possible way 1 People come to you ii Ex ask rst 200 people you see in the Wilk 1 Let39s be honest most of the people eating lunch in the Wilk are freshmen and so they aren39t representative of the entire BYU population b Volunteer response sampling i Individuals select themselves 1 Online pollstelevision pollsmailin polls 2 Bad because we don39t know if they39re biased or not 3 Also bad because people will often only do the polls to say really great things or really bad things ratemyprofessorcom c Quota sampling i Force the sample to meet speci ed quotas 1 Ex 50 girls and 50 guys ii Participants within a subgroup are not selected randomly but rather by convenience or some sort ofjudgement call 5 Why are these bad a Bias don39t know the biases i Sample favors certain outcomes ii Not representative Dunc91 b Impossible to access uncertainty can t make an inference about the population 6 Different examples of bias a Undercoverage disregard a population i Some individuals have no possibility of being selected 1 Ex randomdigit dialing doesn39t really apply to people with only cell phones a Most people ages 1830 only have cell phones so chances are you39d be disregarding them b Nonresponse person doesn39t answer every questionany questions i Ex hang ups on vacation refusal to mail in census forms c Interviewer bias someone might be more open with one interviewer than another i Ex if you39re talking about sensitive subjects you might be more comfortable talking to someone of your own gender ii Answers might be different if the interviewer is rude intimidating or gives subtle clues or gestures d Misleading response i Person may answer dishonestly on sensitive issues 1 Ex did you wash your hands after using the bathroom e Question wording set people up for certain answers i Wording of question leads misleads or confuses 1 Wordydoublenegatives ii Openended questions are really hard to analyze whereas closedended questions limit response options thus forcing people into a box f Question order If you ask a question about happiness in life and then one about debt chances are the answers will be different than if you ask one about debt rst and then happiness 7 How to sample WELL aka Probability Sampling Plans label for all the ways to sample well a Probability sampling i Random samplingselection ensures that the sample is unbiased on average 1 Names in a hat random digit table random number generator randomizerorg b Simple random sampling SRS i Sample of speci ed size chosen such that every possible set of that size n has an equal chance of being in the sample 1 NOT every person has an equal chance but every set ii Problem have to come up with a list of every single person in the population c Strati ed Random Sample i Quota sampling done right 1 Classify population into groups strata based on age gender etc Everyone in each group has a similar characteristic Select SRS from every group Combine SRSs UJN 139 ii Downside must have a list of everyone iii More work than SRS but with less uncertainty d Multistage sampling i Take a sample at each level of hierarchy ex area stake ward member 1 SRS of area a Not from every area like in the strati ed sample b Then for selected areas take an SRS of stakes 2 For selected stakes take SRS of wards 3 For selected wards take SRS of members 4 Combine SRS of members ii Good because it doesn39t require a complete list of population e Cluster sampling i Used when population is naturally divided into groups called clusters 1 A random sample of a cluster is taken 2 All individuals of the selected clusters are included in the sample Poll ALL of group 2 and 11Experimentation a Treatments are imposed i De nition a study design where treatments are imposed on individuals before observing response ii Purpose determine if treatments cause change in response a Experiments can determine CAUSATION iii Why experiment 1 Compared to nonexperimental comparison study a Less chance of confounded lurking variables b Can validly draw causeeffect conclusions b Vocab i Subject individual to which treatment applied ii Factor planned explanatory variable collection of set of treatments iii Treatment experimental condition applied to subjectvaue of factor iv Response variable characteristic measured on each subject v Lurking variables variables that affect response variable but not planned factors vi Control an effort to reduce effects of lurking variables vii Confounding situation where effects of lurking variables can39t be distinguished from effects of factors c For experiments to work i Explicitly describe response variable ii If possible choose homogenous subjects iii Choose treatments to control effects of lurking variables iv Assign subjects to treatments such that groups nearly identical other than treatments no confounding 12Principles of valid experiments a Controlcomparison i Can39t have only one treatment for an experiment ii Control lurking variables by including comparison treatments iii Placebo counts as a second treatment b Randomization i Assigning subjects to treatments randomly helps to neutralize the effects of lurking variables c RepHcann i Multiple subjects per treatment ii DIFFERENT from reproducibility conducting the same study over again and getting the same results d Double blinding i Not necessary but preferred ii Neither the subjects nor those evaluating them know which treatment each subject is receiving Used to prevent the experimenter effect people can tell which treatment they receive based on how the experimenter treats them 13Pitfals in Experimentation a Placebo effect i Response by human subjects due to psychological effect of being treated 1 Those who receive the placebo actually get better 2 Psychological effect is a confounded lurking variable ii Consequence ineffective treatment appears effective relative to untreated subjects iii Solutions 1 Use dummy treatments rather than no treatment as comparison treatment 2 Blind subjects as to which treatment they are receiving b Diagnostic bias i Diagnosis of subjects biased by preconceived notions about effectiveness of treatment 1 Preconception is confounded lurking variable ii Solution 1 Blind diagnosticians doctors 2 Studies in which both subjects and diagnosticians are blinded caed doubleblinded iii Lack of realism 1 Realism often compromised by controlled study conditions choice of homogenous subjects application of treatments a Ex doctors monitor clinical trial patients more than regular people 2 Solution a Awareness of hidden bias your study didn39t replicate reality difference between reality and experiment39s conditions b Admit limitations of experiments iv Hawthorne Effect 1 People in an experiment behave differently from how they would normally behave a Ex food journals diets 2 Consequence inaccurate reporting v Noncompliance 1 Failure to submit to assigned treatment 2 Refusal to follow the protocol of the experiment a Ex they take the medicine but don39t take it the way they are told to 3 Consequence invaid results 4 Noncompliance may undermine an experiment A volunteer sample might solve the problem vi Data ethics 1 How you behave when no one is watching 2 Safety and wellbeing of the subjects must be protected 3 All individuals must give their informed consent before data are collected They must know that there is a chance that they will receive a placebo meaning that they will receive no treatment 4 Individual data must be kept con dential no identity theft