Unit 4 Concept Map
Samples and Surveys
click to edit
Observational Studies and Experiments
click to edit
Samples
Surveys
Observational Studies
Experiments
Sampling Error: An error when using only a section of the entire population to estimate the populations parameters.
Non-Samping Error: Errors that are not a result of sampling but other reasons.
Convenience sampling: a convenience sample is made of people who are easily accessible for sampling.
Voluntary response sampling: A sample that is made up of volunteers.
Under Coverage: A sampling error when a certain group of the population is left out.
A local reported goes to the grocery store to find out what's Duxbury's favorite lunch food option. The result is a caesar salad. This is because the most people who go shopping are mother's for their families. It inaccurately reflects the entire Duxbury population because middle-aged men, teenagers, and kids do not represent the percentage of the sample that they do for the population
Bernie Sanders wants to know people's opinions on free college education. He asks for people to come a conference and give their opinions. The only people that show up to the conference are heavily left winged, early twenty year olds who are suffering from college debt. Since only volunteers showed up for the sample, the sample doesn't properly reflect the entire population's opinion on the issue.
Stratified SRS: The population is divided into groups, also known as strata, that share a common variable. In the strata, random samples are taken.
SRS
Cluster Sampling
When surveys are conducted by a "pie slice" (specific section) of the population.
Can only be used if the population being sampled or surveyed is completely homogenous. If the population is heterogenous, it won't effectively represent the whole population.
I want to take a poll how people feel about the Affordable Care Act. I decide to do the poll via landline. However, many people who are for the Affordable Care Act can't afford landlines, so they are unable to respond to the survey. As a result, my poll concludes that most people, out of the people who took the poll, are against the ACA.
They can be done cheaply and timely, but often don't represent a proper portion of the entire population.
An example could be surveying the top 20% of the class to represent the entire population, if the question is based on religious affiliation.
Randomly surveying a specific number of any random people in the population.
Effective and the desired survey, but are often costly or timely to conduct. They can accurately represent the total population.
Everyone is equally as likely to be selected
click to edit
An example would be random students selected randomly by student ID to answer questions about the length and time of lunch and the food served at school.
Mr. Scozzaro wants to see how effective iLab is for students at DHS. He takes one iLab class from each grade with about an equal amount of college, honors, and AP students. He does a SRS on the classes and looks at the results.
Experimental Units (Subjects)
Studies where existing data is collected based on random surveys. These can't prove a hypothesis because there is a chance that a lurking variable could confound the data, and there are no factors being physically run or enforced.
Control over the variables, and physical running and enforcement of the experiment as opposed to a study where the data is taken and collected from already existing things. Experiments can prove a hypothesis while studies can never factually prove a hypothesis.
Factors
Treatments
Response Variables
The explanatory variables which are the "tests being run." These are the influences being tested on the experimental units(subjects).
The things, animals, or people being tested on throughout the experiment. By observing changes or lack thereof in these units, data is collected.
The different tested groups or categories that the experimental units fall into. They are determined based on the factors (explanatory variables) and their subsections.
The information, qualitative or quantitative data, being collected or observed from the changes or lack thereof in the experimental units.
Lurking Variable
Can confound the data because other variables could alter the data in ways that weren't foreseen or aren't being tested for. For example, the health of a plant could cause some to grow more than others when the variable being tested is the sunlight. This could give inaccurate data and the collected data therefore can never "prove" a hypothesis, instead it can only support it.
Response Bias: Anything in a survey that affects the results
After the Unit 4 Project, Ms. Coleman asks the students "Did you enjoy the project?" Most people are going to say "yes" simply because of who's asking, and not whether they actually enjoyed it or no.
Wording Bias: The wording of a question pushes the answer in one direction.
The President of the Class of 2019 asks fellow classmates "Don't you think that a third iLab day would reduce stress for students?" A less biased question would be "Do you think the benefits of a third iLab day are worth the drawbacks that will result from a third iLab day.
Non Response: When sampling units do not respond to the survey
Mr. Scozzaro sends out a poll for people to take during iLab, but most students do not wish to take it. They therefore don't take it.
Bias: Any systematic failure of a sampling method to represent its population.
An unaccounted for explanatory variable, not being tested in the observational experiment or study, which influences the response variable in an unintended way.
Source: How the bias was made
Direction: What the mistake causes you to overestimate
Example: I want to find out what the favorite Mexican restaurant is for Duxbury High Students. I decide to interview the Crew kids since I was at Snug Harbor after school. Nearly 100% of the Crew members (who are students at DHS) say Moe's is the best Mexican food on the south shore. However, the Crew kids fail to represent Duxbury as a whole since they are some of the only students who would pick Moe's over Chipotle. The source of bias is deciding to interview Crew, while the direction is strongly against the basic trend.
Placebo Effect
A psychological that mentally makes people think something will work, and their behavior and result data will change. For example, if someone thinks that a pain-pill will work for them, their mind might make them think that the pill is working better than it will. This needs to be overcome in order to have accurate data results in a study or experiment (mostly experiment).
Single Blind
Double Blind
The subjects in the study don't know ahead of time that there will be variation in the different groups or tests, and the results will still be accurate. Used to counter the placebo effect.
The subjects in the study don't know ahead of time the variation and the people collecting the data don't know the variation in the different groups or tests. This is a stronger way to counter the placebo effect.
Designed Experiment
Description
Experiment
Factors
Experimental Units
Treatments
Three Principles of Experiment Design
Response Variables
Control: control is for lurking variables that might alter the response. Using a comparative design and making sure the only difference between systems is the treatment administered are ways to control an experiment.
100 people are going to be used to test the effectiveness of a new allergy medication. There will be four medications: one is only sugar, one is the most commonly nationally used allergy medication, one is the proposed dose of the new allergy medication, and the last is a half dose of the new allergy medication. The groups are randomly stratified based on the severeness of their allergies, and there will be 25 people per group. There are 4 possible responses that a subject can provide at the end of the 1 week study: After use of the medication, they now 1.) are experiencing worse symptoms or you had a negative reaction to the medication. 2.) are experiencing no change in their allergy symptoms. 3.) are experiencing slightly reduced allergy symptoms. 4.) They are experiencing highly reduced allergy symptoms. In order to overcome the placebo effect, the subjects will not be told what the medication they are receiving is, how effective it is supposed to be, its price, etc.
Random Assignment: Randomly assigning experimental units to treatments. This helps balance the effects of lurking variables that aren't controlled on the treatment groups by creating equal groups of experiment groups.
Replication: Making sure there is enough experimental units so that differences in the effects can be properly differentiated between groups.
100 Test Subjects
The 4 different [groups]
Different Medication Type
The survey response: Negative, Unchanged, Slightly Improved, Greatly Improved
click to edit
Biggest difference between experiment and observational studies: Treatments are randomly assigned in experiments so you can see cause and effects.