Statistics 115 - Exam 2

clinical trials

medical experiments involving human subjects

nonadherers

subjects who participate but don’t follow the experimental treatment

dropouts

subjects who begin the experiment but do not complete it

• cause systematic bias

matched pair design:

compares 2 treatments by choosing pairs of subjects that are as closely matched as possible. sometimes a “pair” is one subject who gets both treatments (one on each arm, etc)

block

a group of experimental subjects that are known before the experiment ot be similar in some way

• expected to affect the response to the treatment

advantages of block designs

• allows us to draw separate conclusions about each block

• allows more precise overall conclusions

  • control outside variables

institutional review board

reviews all planned studies in advance in order to protect the subjects from possible harm. can require changes to the study or consent form

informed consent

• subjects must be informed in advance about the nature of a study and any risk of harm it may bring

• The subjects should be told what kinds of questions the survey will ask and about how much of their time it will take.

• Experimenters must tell subjects the nature and purpose of the study and outline possible risks.

• Subjects must then consent, usually in writing

assent

children give this, their agreement to participate in the study. must also have their parent(s) or guardian(s) permission that meets the standards of informed consent.

how to reduce bias

use a better instrument to make the measurements

improve reliability

by repeating a measurement several times and using the average result.

double blind experiment

means that neither the participants nor the researchers know which people received which treatment

basic requirement of comparative experiments

equal treatment for all subjects except for the actual treatments the experiment is

comparing

refusals

when people refuse to participate in a study

• can cause bias (if the a lot of a similar group of people refuse, it will become biased)

generalizing

most common weakness in experiments is that we cannot generalize the conclusions widely. Some experiments apply unrealistic treatments, some use subjects from some special group such as college students, and all are performed at some specific place and time. We want to see similar experiments at other places and times confirm important findings.

basic completely randomized experiments

divides all the subjects among all the treatments in one randomization

keys to convincing experiments

randomization, control, and adequate numbers of subjects

confidential

no outside parties have access to individual data

anonymity

subjects’ names are not known to anyone involved in the study, not even to the director of the study. It is not possible to determine which subject produced which data

sources of bias in clinical trials

refusals, nonadherers (taking another drug outside of the one being tested could skew results), and dropouts (If subjects drop out because of their reaction to one of the treatments, bias can result)

potential barriers of informed consent

• possible risks

• long consent forms

• inability to read consent forms

• scared of possible side-effects

to measure something

to assign a number to some property of an individual

variable

result of measurement that takes different values for different people or things

units

used to record measurements/variables

validity

a variable is a valid measure of a property if it is relevant or appropriate as a representation of that property

• a rate is a more valid measure than a simple count

predictive validity

used when we want to measure human personality and other vague properties

errors in measurement

• reduce value of the data

• measured value = true value + bias + random error

reliability

• if random error is small or 0, the measurement is reliable — can still be biased

• means that the result is dependable

• smaller random error = more reliable

random error

repeated measurements on the same individual give different results

round off errors

sum may be slightly off (more than 100%) bc of rounding errors

distribution of a variable

tells us what values it takes and how often it takes those values

bar graph

• better for making comparisons of the sizes of categories

• if there is a natural ordering of the variable, it can be displayed along the horizontal axis of the bar graph, but not easily displayed in a pie chart

• can display entire distribution or display only a few categories

pie chart

• displays the percentage for each category (not count), and only works if you have all the categories (must add to 100%)

trend

long-term upward or downward movement over time (line graph)

seasonal variation

pattern that repeats itself at known regular intervals of time (line graph)