The document discusses different types of experimental designs for analysis of variance (ANOVA), including single-factor, multifactor, randomized block, nested, and BACI (Before-After Control-Impact) designs. Single-factor designs test the effect of one predictor variable, while multifactor designs test the effects of two or more predictors and their interactions. Randomized block designs account for environmental heterogeneity by blocking on environmental factors. Nested designs incorporate subsampling within replicates. BACI designs measure impacts over time by taking measurements before and after a treatment at both impacted and control sites.
The document discusses different types of experimental designs for analysis of variance (ANOVA), including single-factor, multifactor, randomized block, nested, and BACI (Before-After Control-Impact) designs. Single-factor designs test the effect of one predictor variable, while multifactor designs test the effects of two or more predictors and their interactions. Randomized block designs account for environmental heterogeneity by blocking on environmental factors. Nested designs incorporate subsampling within replicates. BACI designs measure impacts over time by taking measurements before and after a treatment at both impacted and control sites.
The document discusses different types of experimental designs for analysis of variance (ANOVA), including single-factor, multifactor, randomized block, nested, and BACI (Before-After Control-Impact) designs. Single-factor designs test the effect of one predictor variable, while multifactor designs test the effects of two or more predictors and their interactions. Randomized block designs account for environmental heterogeneity by blocking on environmental factors. Nested designs incorporate subsampling within replicates. BACI designs measure impacts over time by taking measurements before and after a treatment at both impacted and control sites.
Sampling Designs ANOVA Designs • If you response variables are continuous and your predictor variables are categorical (ordered or unordered)
• The term ANOVA applies to both the Design
and the Statistical Analysis of such designs ANOVA Terminology Treatment: Different categories of the predictor variables that are used. In an experiment it refers to different manipulations that have been performed
In observations, it refers to the different groups
that are being compared ANOVA Terminology Replicate: Within each treatment, multiple observations will be made, and each such observation is a replicate.
In standard designs, the replicates should be
independent of each other, both statistically and biologically ANOVA Terminology Single-factor design: When treatments represent variation in a single predictor variable or factor.
Each value of the factor that represents a
particular treatment is called a treatment level
E.g., Growth responses of an agricultural crop to 4 different
levels of nitrogen or the growth responses of 5 different plant species at one fixed level of nitrogen. ANOVA Terminology Multifactor design: Treatments cover two (or more) factors and each factor is applied in combination in different treatments.
The different levels for each factor may be
ordered or unordered
E.g., Growth responses of an agricultural crop to 4 different
levels of nitrogen and 4 different levels of phosphorus.
This would entail 4 x 4 = 16 different treatment levels
ANOVA Terminology Why Multifactor design? Can two separate single-factor designs not achieve the same?
Advantages:
1. Efficiency in cost, time, and effort to run a single
experiments
2. More importantly, a two-way design allows you to test
both the main effects as well as interaction effects ANOVA Terminology Main effects: The additive effect of each level of one treatment averages over all of the levels of the other treatment.
Interaction effects: Unique responses to
particular treatment combinations that cannot be predicted simply from knowing the main effects E.g., The growth of plants at high levels of N and P may be synergistically greater than what would be predicted from the additive effects of high N and P separately. Single-factor ANOVA The one-way layout: A simple yet powerful design. To compare means among two or more treatments or groups. E.g., Recruitment of Barnacles in an intertidal zone as a function of substrate (say, slate, granite, and concrete) Single-factor ANOVA Disadvantages:
(1) Does not explicitly incorporate environmental
heterogeneity. This is a problem when environmental ‘noise’ is as strong as the ‘signal’.
(2) Treatments are organized along just one factor.
Typically, other factors are present, and interaction terms may be important Single-factor ANOVA Randomized-Block Designs: A modification of one-way ANOVA to include environmental heterogeneity.
Blocks are delineated in which environmental conditions
are relatively homogeneous.
Blocks may be placed randomly, but should be arranged
such that environmental conditions are more similar within blocks than between them.
Best to place only one replicate of each treatment within
Block Orientation: Eg., for three blocks in an elevation
gradient Single-factor ANOVA – Randomized blocks Disadvantages: (1) Statistical cost. If sample size is small and block effect is weak, it is less powerful than a one-way layout
(2) If blocks are too small, you introduce non-
independence due to crowding
(3) If any replicates are lost, data from that entire
block cannot be used
(4) It assumes that there is no interaction between
blocks and treatments Single-factor ANOVA – Randomized blocks Disadvantages:
Therefore some have suggested replication within
blocks. But then that becomes really a two-factor ANOVA.
Replication does help, but is a lot to invest in studying
an attribute that is not of primary interest. Single-factor ANOVA Nested design: Any design in which there is sub-sampling within each of the replicates. Single-factor ANOVA – Nested Designs Nested design: Any design in which there is sub-sampling within each of the replicates.
Note: This does not increase the number of replicates, but
helps in improving precision of the replicate measurements
It also allows you test both variation among treatments,
and the variation within treatments.
Such a nested design can be extended to build a
hierarchical sampling design
E.g., subsamples à replicates à intertidal zones à shores
à regions à sub-continents . . . Single-factor ANOVA – Nested Designs Disadvantages: (1) Often analysed incorrectly. When you treat each sub-sample as independent, for example, and analyse it as a one-way . . .
(2) Difficult or impossible analyse properly.
(3) Most serious is the ‘misplaced’ sampling effort.
Why invest in sub-sampling when it will not qualify as an independent replicate? It is better to invest in getting independent replicates. Multifactor ANOVA Treatments are assigned to two or more factors. Factors may represent ordered or unordered treatments
E.g., say in addition to substrate you were interested
in testing the effects of predatory snails on barnacle recruitment
Predation levels: (1) unmanipulated (2) cage control
(3) predator exclusion (4) predator inclusion
The key element in proper factorial design is to ensure
that the treatments are fully crossed or orthogonal Multifactor ANOVA Multifactor ANOVA It is critical that all treatment combinations are present. Else we end up with a ‘confounded design’
There may be natural covariation between factors in
nature. If you can experimentally break up this covariation, then you can derive powerful inference on the independent contributions of each factor
The key advantage is the power to tease apart main
effects and interactions between two factors
Interactions measure the extent to which different
treatment combinations act additively, synergistically, or antagonistically Multifactor ANOVA Multifactor ANOVA Disadvantage: Treatment combinations can become too large with adequate replication.
For the Barnacle example, with 10 replicates per
treatment, you have 120 total replicates.
As with one-way, the simple two-way design does not
account for spatial heterogeneity
Here again you can use a simple randomized block
design. If you replicate within blocks, you have a three-way design!!!! Environmental Impacts over time: BACI Designs
BACI: Before-After, Control-Impact
A special type of repeated measures design, in which
measurements are taken before and after the application of a treatment
A powerful layout for assessing environmental impact
and monitoring trajectories. Environmental Impacts over time: BACI Designs