- Design of experiment – structured / organized method – determine relationship between different factors affecting process / output of process.
- Includes
o planning experiment
o obtain info from experiment re statistical hypothesis being studied
o making statistical analysis
- Experimental design
o efficient method / planning collection of data
o max amount of information / least amount of work
o directed experimentation generates critical events
o experiments – more structured than sample surveys
§ additional step à treatment
§ sample – select elements from frames / take measurements
§ experimental design – select experimental units / allocate treatments / take measurements
Important Terms and Definitions of Experimental Design
- Observation and experimental studies
o experimental
§ allocate treatment / observe changes in dependent variable / hold other variables constant à ideal
o observational – association
- Experiment
o treatments
o factor – controlled independent variable
§ whose levels are set by researcher
o experimental units
§ smallest division of experimental material to which we apply the treatments
o blocks
§ strata
§ a block is a group of experimental units that are similar in some way that is expected to affect the response to the treatments
· group of homogenous experimental units
o Experimental error
§ chance / uncontrolled factors
§ inherent variability in experimental material
§ lack of uniformity in methodology / failure to standardize experimental technique
§ lack of representativeness of sample to population.
o Blind experiment
§ avoid
· placebo effect
· observer bias
o Natural sources of error in field experiments
§ plant variability
§ seasonal variability
§ soil variability
Three Basic Variables of Experimental Design
- randomization
o allocation of experimental material / order
o eliminate bias
o ensure independence among observations
o criteria for randomization in clinical studies trial
§ each participant – same chance of receiving any intervention
§ allocation – chance
§ simplicity
- replication
o execution of experiment more than once
o average out influence of chance factors
o increases accuracy of estimates of treatment effects
o purpose
§ allows experimenter to obtain estimate of experimental error
§ increases precision by reducing std error
§ can expand base for making inference
- local control / error control / blocking
o more or less uniform experimental units
Design of Experiments in Marketing
- conjoint analysis
- link between customer decision making process & product / service being offered.
- marketing problems
Sample Surveys vs Experimental Design
Parallels between Experimental Designs and Sample Surveys
- random
o choose units from population for sample
o assign treatments to experimental units
- sampling error – minimise by stratification
- blocking to reduce error
- partial grouping useful in cluster sampling
- partial group useful in split plot designs
- analysis
o regression techniques are useful
o ANOVA and ANCOVA are useful.
Study Designs in Medical Research
- Observational studies
o descriptive or case series
o Case control studies (retrospective studies)
§ case and incidence of disease
§ identification of risk factors
o Cross sectional studies, surveys (prevalence)
§ disease description
§ diagnosis and staging
§ disease processes, mechanisms
o Cohort studies (prospective studies)
§ causes and incidence of disease
§ natural history, prognosis
§ identification of risk factors
o historical cohort studies
- Experimental studies
o controlled trials
§ parallel or concurrent controls
· randomized
· not randomized
§ Sequential controls
· self controlled
· crossover
§ External controls (including historical)
o studies with no controls
- meta analysis
Guidelines for Designing Experiments
- Procedure
o recognition of and statement of problem
o selection of response variable
o choice of factors, levels, and ranges
o choice of experimental design
§ potential design factors or nuisance factors
o performing the experiment
o statistical analysis of data
o conclusions and recommendations
-
No comments:
Post a Comment