Friday, April 1, 2011

SG: Topic 1 : Basic Concepts of Experimental Design


 

-          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: