Experimental Design Vocabulary Flashcards

Flashcards for Experimental Design Lecture

Convenience Sample

Easy to gather sample. High bias.

Voluntary Response

People come to you to respond. High bias

Random Sample

Low/no bias. Takes time/effort to gather data.

Census

Gets everyone in the population of interest. Very difficult to get everyone, population changes from beginning to end of data gathering.

Simple Random Sample

A sampling method in which every possible subset of the population of size <n> has the same chance of being selected.

Stratified Random Sample

Break up the population of interest into groups by a variable(the variable picked should have an impact on the responses). Do a simple random sample of each strata with proportionate representation from each group.

Cluster Sample

Works well if the population is broken up naturally into heterogeneous groups. Use random selection to pick 1 or more groups and survey EVERYONE in those groups picked; works well if the population is broken up naturally into heterogeneous groups.

Systematic Random Sample

Pick a random starting point, select every nth person to survey; works well when the population size is unknown.

Bias in Sampling

When your method for gathering data produces results that differ from the parameter of interest.

Undercoverage Bias

Eliminating a portion of the population from the sampling frame. Ex: Only calling people who have landlines typically eliminates younger people who only have cell phones.

Non Response Bias

When the correct people are getting reached, but people are choosing not to respond. Passionate people tend to respond.

Response Bias

When your tool that you’re using to get responses is broken. Leading wording, social desirability, etc.

Observational studies

Used to gather information about a population of interest

Experiment

Used to determine cause and effect between two variables.

Confounding variable

In an observational study, another variable that is associated with the explanatory variable that could also change the response variable.

Treatment

A condition that is assigned to an experimental unit. A control is considered a treatment

Experimental Unit

What is assigned to a specific treatment.

Response Variable

What the researcher records for analysis.

Blinding and why is it important

Allows you to see if treatments are having an impact or if it is the placebo effect.

Direct control and why is it important?

Allows you to control for extraneous variables. This way you can attribute any difference in the response variable to the treatment.

Why is Replication important?

Experimental units might differ in important ways. Having multiple experimental units assigned to each treatment group minimizes those effects.

Placebo

A treatment with the physical characteristics of the other treatments but none of the chemical properties.

Placebo effect

When an experimental unit responds to a treatment due to simply receiving a treatment.

How do you conduct a Randomized Block Design?

Split the experimental units into groups based on a variable that impacts the response. Those groups are called “blocks”. Within each block, randomly assign the EUs to the treatments.

How is a matched pairs design similar to a block design?

In a matched pairs, Experimental units are paired up by a certain characteristic (blocks of size 2).

What happens to sampling variability as sample size increases?

Decreases as sample size increases. The samples all get pulled towards the mean.

Statistical Significance

The difference between a statistic and a population value is too large to attribute to random chance.

What does a Random Sample allow you to do?

Allows you to generalize results to a population of interest.

What does Random Assignment allow you to do?

Allows you to determine cause and effect between two variables.

Describe how to conduct any random assignment/selection:

Label, Select, Link

Label everyone in population with a unique number from 1 - <size of population>
Use RNG to select <sample size> unique values from 1 - <size of population>
Select/survey/assign the people whose numbers correspond to the selected values

When describing bias, what two components must be included:

1. How does the method being produce a sample that differs from the population of interest

2. Will this cause an over or under estimate of the parameter of interest.

How do you discuss a confounding variable

Identify the confounding variable, link it to the explanatory variable, then state that you can’t tell which one is causing the response.

Ex: “People who eat an apple a day tend to be healthier overall.  People who are healthier overall have better dental hygiene.  Therefore, we can’t tell what causes the fewer cavities, people being healthier overall or the apple a day.” 

Why are control groups important?

It allows us to see the actual impact of the other treatments.

How do you select variables to block/stratify?

You want to pick variables that are highly associated with different responses.