4.3.1 Replication
By replication we mean the application of each treatment of interest to more than one
experimental unit (animal, panellist etc.). Replication has two benefits. Firstly, the
population mean response for a particular treatment is estimated by averaging the
observed responses across all the units receiving that treatment. The greater the number
of individual observations contributing to such a treatment sample mean, the greater the
precision with which the treatment population mean is estimated or, equivalently, the
greater the power of the experiment in detecting differences between treatment
population means.
Secondly, two units that receive the same treatment will not necessarily yield the same
response, as variation in response between identically treated units is due to experimental
error. By examining the variation in response between units within treatments, an
estimate of the experimental error variance can be made. This forms a baseline against
which all other apparent differences can be measured, in particular, differences between
the various treatment sample means.
4.3.2 Randomization
The allocation of treatments to units should always be randomized. If not carried out, it
is not possible to determine whether treatment differences are due to treatment or to
confounding by other relevant factors. Randomization is practised to ensure that every
treatment is equally likely to be advantaged or disadvantaged by the selection of units. It
also serves to allow the scientist to proceed as if the assumption of independence is valid.
That is, there is no available (known) systematic bias in how the data are obtained.
4.3.1 ReplicationBy replication we mean the application of each treatment of interest to more than oneexperimental unit (animal, panellist etc.). Replication has two benefits. Firstly, thepopulation mean response for a particular treatment is estimated by averaging theobserved responses across all the units receiving that treatment. The greater the numberof individual observations contributing to such a treatment sample mean, the greater theprecision with which the treatment population mean is estimated or, equivalently, thegreater the power of the experiment in detecting differences between treatmentpopulation means.Secondly, two units that receive the same treatment will not necessarily yield the sameresponse, as variation in response between identically treated units is due to experimentalerror. By examining the variation in response between units within treatments, anestimate of the experimental error variance can be made. This forms a baseline againstwhich all other apparent differences can be measured, in particular, differences betweenthe various treatment sample means.4.3.2 RandomizationThe allocation of treatments to units should always be randomized. If not carried out, itis not possible to determine whether treatment differences are due to treatment or toconfounding by other relevant factors. Randomization is practised to ensure that everytreatment is equally likely to be advantaged or disadvantaged by the selection of units. Italso serves to allow the scientist to proceed as if the assumption of independence is valid.
That is, there is no available (known) systematic bias in how the data are obtained.
การแปล กรุณารอสักครู่..