Section 5 2 EXPERIMENTAL DESIGN EXPERIMENTAL UNITS SUBJECTS

Section 5. 2 EXPERIMENTAL DESIGN

EXPERIMENTAL UNITS, SUBJECTS AND TREATMENTS l l l Experimental Unit – The individuals on which the experiment is being conducted Subjects – Experimental Units that are human beings Treatment – A specific experimental condition that is applied to the units – in an attempt to see if it has an effect

EXPERIMENTAL UNITS, SUBJECTS AND TREATMENTS - EXAMPLES l l l Experimental Unit – 1 acre of land … to have corn planted and harvested later Subjects – A person who will take a test of some sort Treatment – For the above, different brands of fertilizer used… or for the subjects, you might have different types of teaching methods, or background music

FACTORS and LEVELS l l Factor: The Explanatory Variable … For example … A quantity of fertilizer, or drug, or volume of music Level: The amount of the factor being applied to the experimental unit … for example … 100 lb, 200 lb, 500 lb of fertilizer each month per acre. . . 10 mg, 20 mg, 30 mg of a drug … 30 db, 40 db, 70 db of music played during instruction

PLACEBO l l l Placebo: A treatment that does not really have any level of the factor we are treating the subjects with Often referred to a “dummy pill”. . That creates the illusion that a treatment has occurred, when it has not It is a strategy that allows us to compare the response a subject has merely from the fact that they are undergoing a treatment …to the response from undergoing the real treatment

LURKING VARIABLES l l l What are they, and what are we gonna do about them Recall – they were those nasty hidden explanatory elements that might really be at the root cause of some response variable that we cannot account for, know about or control So, a well designed must havce a CONTROL GROUP with randomized assignments to each group(s).

PLACEBO EFFECT Placebo Effect – any dummy treatment that triggers a response form the subject (inanimate objects do not think, and psychologically react to placebos). An expectation for a response on the part of the subject or the experimenter can often create the perception of a observed response. Example- Treating Ulcers – Gastric Freezing – later shown to just be the Placebo Effect

CONTROL GROUP l l l Control Group – The group of subjects or experimental units that do not receive any treatment for the purposes of comparison. (Example: you might conclude the fertilizer made an incredible difference … only to later compare the harvest to the control group and see it is the same. Lurking variable? Maybe incredibly fantastic weather all season long. “CONTROL” is the FIRST BASIC PRINCIPLE of experimental design! – BIG IDEA #1 Without Control – Experiments in medical treatments are ALWAYS BIASED in favor of results showing a positive effect.

RANDOMIZATION BIG IDEA #2 l l l Assigning Subjects to treatment groups Matching of subgroups is helpful – Stratifying can only go so far in controlling Lurking Variables Simplest way to ensure that the Lurking Variables are not responsible for the responses. . RANDOMIZATION … assignment by CHANCE ALONE

CORN A vs. CORN B l l l What if we wanted to see which variety of corn would grow better. We plant Corn A in plot X We plant Corn B in plot Y How can we tell if the corn grew better due to the brand of corn? … or the soil, sun, etc. impacting the two plots X and Y? RHETORICAL … you can’t!

30 RATS l l l Number all the rats 01 through 30. Use a table of random digits or a computer to generate random values until 15 rats have been selected. Assign them to group A Assign the remaining to group B … or alternate group A then group B then group A, etc.

RANDOMIZED COMPARATIVE EXPERIMENTS l l l Randomization of assignments assures that experimental units in each group are relatively similar in all respects before treatment Comparative Design ensures that influences other than the treatment operate equally on both (all) groups. Therefore differences in response must be due to the treatment OR the random effects of chance

EXPERIEMNTAL UNITS & n l l l BIG IDEA #3 – Use a big enough n – sample size – so the effects of the variation even out. The effects of chance do average out over the long-run – and with larger samples If you only use ONE or a FEW units, then the effects of chance are magnified, and the link to the treatment is ore uncertain

PRINCIPLES OF EXPERIMENTAL DESIGN - SUMMARY l 1. CONTROL l 2. RANDOMIZING l 3. REPLICATION

STATISITICAL SIGNIFICANCE l l l Statistical Significance – When an observed effect is so large that it would very RARELY occur by chance Are we saying “IMPOSSIBLE” to occur … NO But YOU the student of statistics must embrace new interpretations of “POSSIBLE”. . . Or “WILL HAPPEN” Flipping a coin right now in front of you 20 times inb a row HEADS … leads you to conclude that SOMETHING IS FAKE … rather than a RARE EVENT ACTUALLY JUST HAPPENED! But to be clear – WE NEVER PROVE anything in Statistics

COMLETELY RANDOMIZED l Completely Randomized – All experimental units are assigned at random among all treatment groups Group 1 20 Houses Random Assignment Treatment 1 Meter Group 2 20 Houses Treatment 2 Chart Group 3 20 Houses Treatment 3 Control Compare electricity use

DOUBLE-BLIND l Double-blind - Neither the subjects nor the experimenter are aware of what treatment is being received by the subjects l Example: I administer 3 types of pills to patients in three groups Treatment A, B and the Control Group. I do so by having someone else code the pills, and I record which patients revived which coded pill.

LACK OF REALISM l Lack of Realism – occurs in an experiment where the conditions, environment or situation does not realistically replicate the conditions that we want to study l l Example: Do YOU or I behave realistically when you know you are being OBSERVED or EXPERIMENTED ON? Example: High Center Rear Brake Lights. . I’ll call it the novelty effect – a lurking variable

MATHCED PAIRS DESIGN l l l Matched Pairs Design – An experimental design in which two units are blocked together to receive the two different treatments. A subject might receive both treatments, one after the other. Several subjects might be involved to allow for A follows B on half, and B follows A on the other half. Before and After experiments are also an example. Rate sweetness … FREEZE. . Rate sweetness again

BLOCK DESIGN l l Block – a group of experimental units or subjects that are known before the experiment to be similar in some way that is expected to affect the response to the treatments Block Design – the random assignment of units to treatments is carried out separately within each block