View
214
Download
1
Category
Tags:
Preview:
Citation preview
ExperimentImposing treatments on the
subjects in order to compare the results.
Explanatory variables are called factors and specific values of the explanatory variable are levels. These determine the treatments.
The objects on which the treatment is imposed on are called experimental units (human subjects).
Does the type of lighting or the type of music in a dentist’s waiting room have any effect on
the anxiety of a patient?• Types of music:
– pop
– classical
– jazz
• Levels of brightness– low
– medium
– high
• What are the factors?
• What could a possible response variable be?
• How many treatments are there?
• How many levels are there of each factor?
Type of music & Brightness level
3
9
Blood pressure, ...
Does the increase in the explanatory variable CAUSE the increase in the response variable?
YearNumber of Methodist Ministers in New England
Cuban Rum Imported to Boston (in barrels)
1860 63 8,376
1865 48 6,406
1870 53 7,005
1875 64 8,486
1880 72 9,595
1885 80 10,643
1890 85 11,265
1895 76 10,071
1900 80 10,547
1905 83 11,008
1910 105 13,885
1915 140 18,559
1920 175 23,024
1925 183 24,185
1930 192 25,434
1935 221 29,238
1940 262 34,705
The Beauty of ExperimentsJust because two variables have a relationship, that
doesn’t mean one causes the other – there is often a confounding variable at play!
BUT…long term well-designed experiments CAN be used to imply CAUSATION between the explanatory variable and the response variable.
Observational studies/ surveys can NOT!
http://tylervigen.com/
In which option can we conclude that the change in the explanatory variable CAUSES the biggest
change in the response variable?• Option 1: Students choose which group to join
• Option 2: Students are randomly assigned to a group.
• Students in Red Group– Do jumping jacks for two minutes
– Then measure heart rate
• Explanatory variable: type of activity (jogging or jj)
• Response variable: heart rate
Designing Good Experiments • Students in Blue Group
- Jog in place for two minutes- Then measure heart rate
3 Principles of Good Experimental DesignReplication--consistency to many subjects
Randomization--randomly assign subjects to treatment groups
Control/Comparison—having at
least two groups that can be
compared
“Differences in the response variable between groups, if enough to rule out natural chance variability, can then be attributed to the manipulation of the explanatory variable.” This will allow determination of cause and effect.
Ways to Randomize• Each subject draws out
of a bag a colored chip - each treatment is a different color
• Each subject is assigned a number - first set of random numbers produced go to one treatment and the rest to the other
• Each subject rolls a die - odds go to one treatment, evens to the other
• Each subject flips a coin - heads go one treatment, tails to the other
Control group--receives standard/traditional treatment OR no treatment at all OR a
Placebo -receives no active ingredient but subjects believe they are receiving treatment
Single Blind: subjects don’t know which treatment they receive
Double Blind: subjects and evaluators are “blind”; only the researcher has the “key”
Other Experimental Vocabulary
Completely Randomized Design• Randomly assign a treatment to each experimental
unit• The number of units assigned to each treatment is
as equal as possible• Randomization is expected to spread any
differences among units equally across all treatment groups
• Any significant difference in the two groups’ responses can be attributed to treatments used – therefore there are no confounding issues
Is weight training good for children? If so, is it better for them to lift heavy weights for a few repetitions or moderate weights a larger
number of times?
43 volunteers
14--Heavy load group
15--Moderate load group
14--Control group – no weights
Measure and compare muscular strength & endurance
Include: •Subjects•Random assignment•Explanatory(treatments)•Response
Randomly assigned to 3 treatment
groups
Block Designs to Reduce VariabilityBlock Design--divide units into
groups (blocks) in which the units in each block are similar to each other. Within each block randomly assign treatments (do multiple CRDs).
Block if you have reason to believe certain groups will have different results.
Matched-Pair Design—blocking on a unit: randomly assign either two matched units (identical twins) the treatments OR the same individual receives both treatments in random order
Blocking reduces variability.
Randomized Block Design
43 volunteers blocked by age
21-- Children aged 5-10
7--Heavy
22-- Children aged 11-16
Measure & compare between groups
7--Moderate
7--Control
7--Heavy
8--Moderate
7--Control
Measure & compare between groups
Block by age because we believe younger children might have different results than older children
Randomly assigned to 3 groups
Randomly assigned to 3 groups
Matched Pair Design
43 volunteers
21—Heavy load for six weeks, rest two weeks, moderate load for six weeks
22 – Moderate load for six weeks, rest two weeks, heavy load for six weeks
Measure increases in strength & endurance & compare between groups
Randomly assigned to
2 groups
Quitting Smoking w/Nicotine PatchesRecruited 240 smokers (volunteers) at Mayo Clinic
from 3 large cities
Randomly assigned 22-mg nicotine
patch or placebo patch for 8 weeks.
All attended counseling before, during, and after.
Double-blind
After 8-wk (1 yr), 46% (27.5%) of nicotine patch group quit smoking and 20% (14.2%) of placebo group quit.
Quitting Smoking w/Nicotine Patches
• What are the experimental units?
• What are the treatments?
• What was the explanatory variable?
• What was the response variable?
• How was randomization applied?
• How was control applied?
• How was replication applied?
• Is this an experiment or an observational study?
• How would you summarize the results of this experiment?
• 240 volunteers• Patches• Type of patch• Whether or not they quit• Assigned patch
randomly• Placebo patch • About 120 in each group• Experiment
• The nicotine patch worked!
Which type of experiment?
• A baby-food producer claims that her product is superior to that of her leading competitor, in that babies gain weight faster with her product. As an experiment, 30 healthy babies are randomly selected. For two months, 15 are fed her product and 15 are feed the competitor’s product. Each baby’s weight gain (in ounces) was recorded.
• Completely Randomized Design • Randomly assign babies to treatments.
Have 15 red and 15 blue chips in a bag and draw one for each baby. Reds get her product, blues get the competitor’s product.
• Or …Assign each baby a number 01-30. Generate 15 distinct random numbers in this range to get her product. The rest get the competitor’s product.
• Compare the weight gain of the babies after the two month period.
Which type of experiment?
• Is the right hand of right-handed people generally stronger that the left? Paul Murky of Murky Research designs an experiment to test this question. He fastens an ordinary bathroom scale to a shelf five feet from the floor, with the end of the scale projecting out from the shelf. Subjects squeeze the scale between their thumb and their fingers on the top. A scale, which reads in pounds, will be used to measure hand strength.
• Matched Pairs Design
• Randomly assign ½ of the people to test right and then left hand, and the other ½ to test left hand first, and then right hand.
• Make sure the participants cannot read the scale so they don’t influence themselves into trying to “top their score”.
• Compare the differences in hand strength.
Which type of experiment?
• An agronomist wishes to compare the yield of five corn varieties. The field, in which the experiment will be carried out, increases in fertility from north to south.
• Block design• Block the field on
location since fertility increases from north to south. (each color is one block)
• Randomly assign each plot within the block to 1 of the 5 corn varieties (1-5).
• Compare the yield of corn from each of the plots of land.
Randomization in ALL Studies
• In Observational Studies
– Randomize selection of subjects
• In Experiments
– Randomize assignment of treatment
Using Random number table to randomize assigning treatments
• 15487 45195 56420 02314 41265 03798 23185 15770 21468 02172 39741 01468 15647 04841 54970 32670
• Suppose you have 3 treatments for 30 subjects • Option 1: Assign subjects a number and first 10 numbers chosen
get treatment A, second 10 get B and the rest get C.
• Option 2: Assign the treatments a number or range of numbers (A is 1-3, B is 4-6, C is 7-9, ignore 0) and each subject gets the next treatment that comes up. Stop when there are 10 in each treatment group.
Recommended