Ch4 - Designing Studies

population

entire group of individuals we want info abt in a study (ie all Wellesley students)

census

collects data from every individual in the population

sample

is a subset of individuals in the population from which we actually collect data (ie 10 students)

5 types of samples

1. connivence sample

2. systematic sample

3. simple random sample (SRS)

4. stratified random sample (alt to SRS)

5. cluster sample

a convienence sample

is a sample that uses individuals in the population who are easiest to reach (ie all ppl in caf)

a systematic sample

setting a system such as select every nth individual to take a sample

simple random sampling (SRS)

a sample that is chosen where each individual in the pop has the same chance of being selected for sample (selection & assignment r same)

stratified random sample (alt to SRS)

classifies pop into groups of similar individuals (strata), chooses a separate SRS in each stratum, combines SRS’s to form sample (random in selection strata’s & randomly assign strata’s to treatment?)

blocking

similar to stratified random sample, blocking involves dividing subjects into subgroups to control variables for research (& random ASSIGNMENT of units to treatment is carried out separately in each block)

a block

a group of units known before experiment to be similar AND expected to effect response to treatment

bias

results in consistent underestimation or overestimation of a value

• ie how much hw? only survey lib students = overestimation

• occurs due to a problem in experiment design

voluntary response sample

consists of ppl who select themselves by responding to a general invitation (=bias)

random sampling

uses a chance process to determine members of a pop included in the sample (most accurate/least bias)

strata

a group of similar individuals (sample)

cluster sample

classifies pop into groups of individuals near each other (clusters), chooses clusters using SRS, ppl in chosen clusters form sample (cluster gets 1 assignment)

clusters

groups of similar ppl

inference

a conclusion drawn abt a pop based on sample data; what we know abt the sample is applied to the pop.

undercoverage

occurs when some members of the pop cannot be chosen in a sample

nonresponse

occurs when an individual chosen for the sample cannot be contacted or refuses to participate

response bias

when ppl consistently answer inaccurately due to the race, age, gender, ethnicity, or behavior of the interviewer

ways to randomly select

• random number generator

• hat method

• random digit table

an experiment

deliberately imposes some treatment on individuals in order to observe their responses (cause & effect)

an observational study

observes individual’s & measures variables of interest but does not attempt to influence the response

experimental units

individuals in experiment

subject

humans in experiment

explanatory variable

a variable/factor/independent variable that is manipulated to observe its effect on the response variable in an experiment (1 variable: type of feed, 2nd: hours of sunlight)

factors

are called explanatory variables (1 variable: type of feed, 2nd: hours of sunlight)

levels

the values of a factor (oz of feed, strength of UV)

treatement

the specific experimental condition applied to the units (same as the level when 1 variable)

response variable

end result, dependent variable, what u measuring

control group

receives an inactive treatment or existing baseline treatment

A well designed experiment has:

4 things

1. comparison.

2. random assignment.

3. control.

4. replication.

A well designed experiment has comparison.

use a design that compares 2 or more treatments

• control group: no treatment/baseline

A well designed experiment has random assignment.

use chance to assign experimental units to treatment

• this helps create roughly equivalent groups of experimental units by balancing the effects of other variables among treatment groups

• AND can only determine cause and effect (vs observation = infer)

A well designed experiment has control.

keep other variables that might affect the response the same for all groups

A well designed experiment has replication.

use enough units in each group that any diff in the effects of treatment can be distinguished from chance differences between groups

confounding variables

2+ factors that both influence response variable and are related, and therefore cannot be distinguished from each other leading to a false association

in a completely random design

only the experimental units have to be assigned to the treatment completely by chance (not to do w/ selection)

placebo

a neutral treatment that has no “real” effect on the dependent variable

placebo effect

the response to a dummy treatment (some ppl will respond just bc they took a “pill”)

statistically significant

an observed effect that is so large it would rarely occur by chance alone (necessary for replication) (if not sig, the difference wasn’t large enough to rule out random chance as a plausible explanation)

double blind experiments

studies where neither the subjects nor the people who interact w/ them and measure their response know which treatment is being received (placebo or real)

design of an experiment diagram (SRS)

experimental units → random assignment »group 1 / 2 » treatment 1 / 2 » compare results

design of a randomized block design

experimental units »group 1 / 2 » random assignment » treatment 1 / 2 » compare results

in a randomized block design

random assignment of units to treatment is carried out separately in each block

a matched pair design

is a random block experiment which each block consists of a matching pair of similar experimental units (only for 2 treat)

• either 1 individuals RA to both treatments

• or 2 very similar units RA to treatment

independent variable

the variable that is manipulated in an experiment to observe its effect on the dependent variable (imovable, consistent) ie time

dependent variable

the outcome measured in an experiment that is affected by the independent variable (changes) ie speed

rows vs columns

left to right (horizontal) vs up and down (vertical)