4b: 3.5-3.7 Experiments Flashcards
How can you tell if a study was an experiment and not just an observational study?
Did they “force” some kind of treatment on the individuals? Did they randomly split them into multiple groups?
When is the only time you can conclude that one variable is causing a change in a different variable?
If it was an experiment (treatments given to random groups)
What is a response variable?
In an experiment, this is what is actually measured and compared at the end of the experiment. Examples: SAT Scores from each group, the percentage of cows that were killed within each group, etc.
Why is it advantageous to have more individuals in a study?
More data = less variation.
Note: more people will make any conclusions/predictions more precise but it will not affect the accuracy. If your sample/experiment is bad, having more people won’t help.
In an experiment, why do we use random assignment when splitting into treatment groups?
To create roughly equivalent groups for each treatment (as similar as possible). This will also help to eliminate some potential confounding variables. Example: “smart” students will score better on the SAT regardless if they took a prep course. However, since we randomly assigned the students to take a prep course or not, just as many “smart” students should end up in each treatment group.
In an experiment, why would we want to use blocking?
If one specific variable will strongly affect the results, we might want to split the subjects into multiple groups based on that variable before we randomly assign treatments (rather than just relying on the random assignment to do the job). Example: we are sure the Seniors will do better than Juniors so we will split up the groups by grade level first and then randomly choose half of the seniors to take the prep course, etc…
What does “blind” mean in an experiment?
Someone doesn’t know what treatment is being given. In a single-blind experiment, the subject is unaware which treatment they received. In double-blind, the experimenter also doesn’t know which subjet is getting which treatment.
Why might we use blinding in an experiment?
To reduce the placebo effect. Double-blind would help eliminate any possibility that the experimenter will try to manipulate (consciously or subconsciously) the results.
What is a control group and why would we use one?
A control group is a treatment group that essentially gets no treatment or a placebo. This will allow us to have something to compare the other treatment(s) to. If you have two different treatments, you don’t absolutely need a control group because you can compare them to each other. Example: the cows that received no spots on their bums were the control group.
How do you explain how to randomly assign the subjects to treatment groups?
(Look Real Close)
Put all their names in a hat and mix well. Randomly choose (X) names and give them (treatment A). Repeat for each treatment.
If the subjects aren’t human:
Assign each (individual) a number. Randomly select (X) numbers without replacement and give the individuals whose numbers are picked (treatment A). Repeat for each treatment.
How can you tell if an experiment uses matched pairs?
What are they comparing at the end? If you would take a difference of a bunch of pairs of data, it is matched pairs. Most common: compare someone with themselves while “on” each treatment. Example, give some people caffeine one day and a placebo the other day. Compare the average change in heart rate for each subject on caffeine and not.
What is a confounding variable?
Some other factor (not the one being tested) that might also produce the same result as the supposed explanatory variable. If explaining, it should be something that also goes along with the explanatory variable. Example: if you just ask who took a prep course, it could just be that these are more motivated students and would score better on the SAT even without the prep course. We can’t say that the prep course increased scores.
In statistics we don’t prove anything, we gather evidence and make a conclusion. What kind of evidence is considered statistically significant?
Any result that would be rare by chance alone. Later in the course, we will discover that the most standard degree of rarity is 5%. So if a certain result would happen less than 1 in 20 by chance alone, it is good evidence for something else.
When given the results on a simulation (usually a dotplot), how can you tell if some specific result is statistically significant?
Approximate what percent (how many dots of the total) are at or more extreme than the result in question. If there aren’t many (<5%) then you have good evidence.