Вы находитесь на странице: 1из 13

Design of Experiments

An experimental design is a plan for assigning experimental units to treatment levels and the statistical analysis associated with the plan. The design of an experiment involves a number of inter-related activities. The Primary goal of an experimental design is to establish a causal connection between the independent and dependent variables. A secondary goal is to extract the maximum amount of information with the minimum expenditure of resources.

1. Formulation of statistical hypothesis. 2. Determination of the treatment levels to be manipulated (independent variables), the measurement to be recorded and extraneous variable to be controlled. 3. Number of the experimental units required and the population from which they will be sampled. 4. Statistical analysis of the results.

Sample Survey and Experiments


Sample survey
characteristics of some population. Focus on sampling of individuals from the population Perspective (the subjects explanatory level is identied rst, then the outcome or response of interest is observed) and retrospective (the subjects outcome or response of interest is observedis identied rst, then the explanatory level).

Experiment discover the effect of applying a stimulus to subjects.


Focuses on formation of comparison groups and allow conclusions about the effect of the stimulus to be drawn.

Similarities
Carefully drawn blueprint or design Both involves statistical consideration before starting.

Differences
Experimenter actively intervenes with the experimental subjects. Survey researcher passively observe or record the responses of the survey subjects. Experimenter goal is to find out relation between independent and dependent variables. Response of the dependent variable by changing independent variable (at different levels). Survey goal is to describe the population. No manipulation is done in the survey variables. work on opinion polls, market research and surveys of social, medical, environmental and other issues

Comparison groups and randomization


In studying the effect of independent variable on the dependent variable, extraneous variable has to under consideration. Cofounded variables: when the effect of independent and extraneous variable can not be separated. The sample survey the studies are cofounded. The cofounded variables can be separated by selecting the appropriate design.

Use of comparison groups- giving the treatment to one of the two equivalent groups of subject and withholding it from the other group. Matching- subject are matched on age gender, race or other characteristics. Randomization- eliminate bias, remove effect of extraneous variable.

Single blind and double blind experiment


Ways to avoid the bias. In drug trials
Single blind_ The subjects are blinded- not informed about the medication
control group may lose hope, and treated goup may become +, because of some expectation in the results.

Double blind_ both the treatment provider and the subject are unaware of the treatment.
Experimenter may become + and the recording of result may be errored.

Random assignment and sample size


Random sampling without replacement procedure. Randomly assigns the subject to group. If one were to assign 60 subjects to 3 groups, the 1st 20 will be assigned to the 1st group, 2nd 20 will be assigned to the next group and the remaining 20 will be assigned to the 3rd group.
Chances of extraneous variables Large no of samples should be there for equal distribution of the extraneous variable throughout the group.

Blocking and extraneous variables


Assignment of subject to any group is CRD (control randomized design).
Sample random sample survey By randomization effect of extraneous variable is deleted Moreover the sample size should be large

Blocking is another method to eliminate the extraneous variables. Blocking: directly taking the extraneous variable into account in the design.

Example
Effect of different diets on weight loss. Subjects are blocked or grouped in different initial weight catagories. Within each block subjects are then randomly assigned to different diets. Blocking guarantees that each diet has subjects with the same distribution of initial weight. Randomization can not totally guarantee this. Similar to stratified sampling. The experimental design that uses blocks to control the effect of extraneous is RBD (randomized block design). Randomization within the group.

Limitations of experiments
Sometimes we study an character in some people, and it is studied in some limited area.
Does this mean that this is the character of the population.
No

Experiments take place in highly controlled and artificial environment.


Condition and environment under which the experiment is conducted is also very important. Time, place, duration, other conditional factors.

Sometime cause and effect in an experiment is difficult to analyze.


Carefully designet experiments

The statistical principles of experiments can sometimes be in conflict with our cultural and ethical standards.
If the direct potential hazard is there to humans with respect of drug testing it is unethical to test it. Voluntary, not pressurizing.

Вам также может понравиться