Statistical Sciences 2244A/B Lecture Notes - Lecture 2: Confounding, Dependent And Independent Variables, Frankenmuth, Michigan
22 May 2018
School
Department
Professor
Stats 2244
Lecture 3
Study Designs and Issues
- Your sampling design is how you choose the individuals in your sample and study designs are
what you do to those individuals
Issues with observational studies
- Association bw our explanatory variables and response variables is not necessarily a cause and
effect relationship
o ASSOCIATION IS NOT NECESSARILY CAUSATION
o Just bc there is a relationship bw 2 variables does not mean that one causes the other
- Confounding variables
o We don’t know which variable is responsible for the results that we are getting
o The confounding variable is another variable other than our factors of interest
o For there to be a confounding variable, there must be a clear change (increase or
decrease) of the 2 variables at the same time
o We cant separate the influence of the confounding variable vs our factors of interest on
the response variable
Principles of experimental design (these are the 3 big principles)
- Replication:
o Repetition of a treatment on different units within a study (different from repeating the
study)
o You have more than one unit (individual or object) in our treatment groups
o This is important bc it allows your results to be more representative of the population
(bc there is variation within the samples)
o So its important bc it allows us to account for variation bw individuals
- Randomization
o Random assignment of units (individuals) to treatments OR the order of treatment
experienced
▪ Treatment groups are homogenous on average
▪ more often than not you end up with groups that are similar (this is caused by
randomization)
▪ note: the different treatment groups are going to be very similar this doesn’t
mean that each individual is similar)
o Don’t confuse this with random sampling
o Question: there are 16 males and 8 females. Males, on average are taller than females. If
we repeat the randomization process many times, what do you think the typical
difference in mean height (mean of group 1 - mean of group2 ) bw the groups will be?
▪ a) difference <0
▪ b) different about = 0
▪ c) difference >0
• answer: B
o Question: what do you think the typical proportion of males (Pmales in group1 - Pmales in group2)
bw the groups will be?
▪ a) difference <0
▪ b) different about = 0
▪ c) difference >0
• answer: B
o this eliminates or at least reduces confounding variables
find more resources at oneclass.com
find more resources at oneclass.com
Document Summary
Your sampling design is how you choose the individuals in your sample and study designs are what you do to those individuals. Association bw our explanatory variables and response variables is not necessarily a cause and effect relationship. Confounding variables: association is not necessarily causation. Principles of experimental design (these are the 3 big principles) Worry about researcher bias (could also apply to observational studies: when the researcher unintentionally interprets or records data based on their expectations or wants, subjects changing behavior bc they know they are being observed. What"s wrong here? affects the way people rate breakfast cereal. The same cereal is put in three packages, (i) a white. Color of the cereal box we don"t know whether the color of recruited to participate in the study. Each volunteer was given one box of cereal to try for the following week. Each cereal box was wrapped in brown paper when distributing to the participants (who unwrapped it at home).
