UNIT 3 - sampling and experiments Flashcards
Can you stratify in an experiment?
NO. stratification is a sampling method, blocking is method used in experiments. They are similar ideas.
How are we proving causation in experiments and obs studies?
No causation in a study, maybe association or correlation. ONLY IN EXPERIMENTS TO YOU TALK ABOUT CAUSALITY.
When your sampling frame is different from the population, then you risk ____
undercoverage
What is wording bias
A type of response bias, When the wording of the question impacts response to it. (type of response bias)
Systematic, how do you find the N for every nth subject, and then how do you proceed?
TOTAL POP/SAMPLE SIZE= your n (round down). Then use RAND INT to Randomly choose first. RANDINT(1, n). And then take every nTH.
What is the standard sampling method? (the gold standard)
A Simple Random Sample (SRS) is our standard. Every possible group of n individuals has an equal chance of being our sample. That’s what makes it simple. Put the names in a hat.
How is Blocking in an Experiment Similar to Stratifying in a Sample?
The two are similar because they divide the subjects into homogenous groups where the subjects are all similar (these traits were already present in the population)
What is BIAS in sampling?
A systematic FLAW in your method.
How are we manipulating the environment differently in experiments and studies?
No manipulation or treatments in an observational study. You only manipulate environment in an experminet.
What is “control?”
You want to control the environment as best as you can so that the only difference between groups is the treatment, and the treatment only. Everything else should be similar.
Give example of randomized block design for a new anxiety med vs placebo with 100 volunteers (60m and 40f)
Block by gender. Randomly assign 30m to new and rest placebo. Randomly assign 20w new and rest placebo.
Name types of bias
undercoverage, non response, response, voluntary
What is the placebo effect?
When those who get the placebo show improvements, or show the effects of the treatment. This often happens to up 20% of participants!
what is the best way to reduce bias?
randomness and good sampling methods.
How are we making inferences differently in experiments vs studies?
In observaional studies, you make and inference about the population, in an experiment you make an inference about a treatment.
What is “replication?”
Having enough subjects. You don’t want to test fertilizer on just one plant.
Give an example of matched pair design for comparing a new blood pressure medication to an older version?
Have subjects use the new medication for a month and the old one for a month and compare. Be sure to randomize which month and blind.
Who can be blinded? ( two groups)
- Subjects (and dog owners..). The poeple getting treatment and 2. administrators. Those delivering treatments and assessing effectiveness of treatments.
What is a weakness of a SRS?
Suppose you want a sample of 50 high school students, with an SRS, although unlikely you could get “all freshmen” which wouldn’t be representative.
A 4 year high school of 2000 students, sampling 40 high students: Describe a cluster sample
Imagine that all of art classes have 10 students and they are mixed with fr, so, jr and srs… You would randomly choose 4 classes and survey everyone in each of the 4 classes.
Why does it make sense to double-blind an experiment?
It reduces bias in an experiment. If subjects don’t know what treatment they’re receiving, they won’t change their habits based on that knowledge. If evaluators don’t know which treatment each subject is receiving, they won’t bias the true results based on the results they expect to see
How are statistics and parameters different?
A statistics is a numerical description of a sample, and a parameter is a description of a population. The difference between these is called ERROR (sampling error).
A 4 year high school of 2000 students, sampling 40 high students: Explain how stratifying has “impossible groups”
You couldn’t get all freshmen in your sample
Suppose you want to see the relationship between gender and candy preference in squirrels. How may you do a stratified vs cluster sample
STRATIFIED: You can split the list of all of the squirrels in your neihborhood by gender and randomly select 20 males from th list of all of the males, and then 20 females (strata) from all of the females. CLUSTER: you can randomly choose to 5 different trees and survey all of the squirrels in those trees, assuming that there are 4 squirrels living in each tree (clusters, the trees have both M and F).
What is difference between non response bias and undercoverage?
You may ask someone to take a survey, they may say no. They may feel differently than the people who decide to take the survey. In this case, that is non-response bias. Undercoverage happens when you didn’t even ask some people to take the survey. The people you didn’t even ask might feel different.
Is it always better to do a census or to sample?
depends on the availablility of the data. If the you want to look at SAT vs GPA, you may easily be able to get all of the school’s data and do that study (a census). If you have to go out and get the info, you may want to take a sample to save time and energy.
How is BIAS different from SAMLING error
Bias is a systematic flaw in your sampling method. Sampling error is always present even with the best methodology.
What is the difference between single-blind and double blind?
Single blinding is when all individuals in either one of the classes are blinded; double-blinded is when everyone in BOTH classes are blinded. Classes are: subjects and the other is treatment givers + evaluators…
name 2 differences between observational studies and experiments
- Experiments can prove causation (studies can’t) . 2. In experiments, you assign treatments (studies you just watch)
What is the purpose of matching?
it isolates the differences between subjects so help see the impact of the treatment.
Give example of a matched pair design study for fuel efficiency
Testing fuel efficiency of different gasolines. Subjects use both fuel A for a month and fuel B for a month and compare, based on their driving habits and vehicle, which was more efficient. BE SURE TO RANDOMLY CHOOSE WHICH ONE GOES FIRST FOR EACH SUBJECT.
What is the difference between response bias and nonresponse bias?
Response is when the person’s response is influenced by the question or questioning method (like if a parent asks if you use drugs, as opposed to a friend… there is only one answer to this, but one might respond differently to them), non response is is when the people who don’t respond might have different opinions/views than the people who did.
What is a sampling frame?
It is the frame from which you get your sample. For instance, if you call people the frame would be “people with phones”
A 4 year high school of 2000 students, sampling 40 high students: Describe a stratified sample
Stratify by year. Randomly choose 10 FR, 10 SO, 10 JU and 10 SENIORS
Example of response bias
A teenager goes to the doctor’s office with their parents. The doctor asks the teen if they vape. The teen may say “no” because their parent’s are there, even though they do vape.
What is the main purpose of a placebo ?
To blind the subject that is being experimented on to avoid influence to the given variable therefore altering the response variable . When people think they’re getting help, they often improve anyway..
4 INGREDIENTS TO EXPERIMENTS
Compare, control , randomization, replication (and BLOCKING when you need to)
Example of undercoverage
You only ask people who go to Home Depot about their views on school lunches.
What is a simple random sample?
put all of the names in a hat. every group is possible. pull the numbers
When we say “statistics vary” or the “variablility of statistics” are we talking about data from an individual?
NO… we are stating that summaries of samples (statistics) will vary from sample to sample. Statistics from one sample will differ from statistics from another sample and they will also differ from the parameters. The distance your statistic is from the parameter is called the ERROR.