ICAR AICE-JRF/SRF Forestry Research Methods & Experimental Design: practice questions

1 · Scientific method and research problem formulation · medium

Why can a well-designed manipulative experiment support a causal claim more strongly than an observational study across sites?

1. Experiments always use more replicates than any observational study can
2. Randomisation breaks links between treatment and unmeasured confounders
3. Observational studies cannot be analysed with statistics
4. Experiments do not require any measurement of the response

2 · Scientific method and research problem formulation · hard

Two treatments are to be compared with σ = 4 cm, smallest important difference δ = 3 cm, α = 0.05 (z = 1.96) and power 0.80 (z = 0.84). Using n ≈ 2(z₁ + z₂)²σ² ÷ δ², how many replicates per treatment are needed?

1. 14
2. 28
3. 21
4. 56

3 · Scientific method and research problem formulation · hard

Which practice best reduces the risk that a research question is unanswerable with the planned study?

1. Increasing the number of response variables measured
2. Pilot data and a power analysis before committing to the design
3. Choosing a significance level of 0.10 rather than the customary 0.05
4. Deciding the analysis only after seeing all the data

4 · Factorial, split-plot and incomplete block designs · medium

Alpha designs are popular in tree breeding trials mainly because they are:

1. Complete block designs with one tree per family in each block
2. Designs that need no randomisation of entries
3. Resolvable incomplete block designs flexible in size
4. Systematic designs for spacing comparisons

5 · Mixed-effects models and repeated measures · hard

Two models differ only in their fixed effects. To compare them by AIC or a likelihood ratio test you should fit them by:

1. Maximum likelihood
2. Restricted maximum likelihood
3. Ordinary least squares on plot means
4. The method of moments

6 · Mixed-effects models and repeated measures · medium

Best linear unbiased predictions (BLUPs) of family effects in a breeding trial:

1. Equal the raw family means in every possible case
2. Ignore the number of trees per family
3. Are used only for fixed effects
4. Shrink extreme family means towards the mean

7 · Mixed-effects models and repeated measures · hard

Model A has log-likelihood −250 with 5 parameters; model B has −246 with 8 parameters. Which has the lower AIC?

1. Model A: −490 against −476 for B
2. Model A: 500 against 492 for B
3. Neither: both equal 509
4. Model B: 508 against 510 for A

8 · Multivariate methods: PCA, cluster analysis, ordination · hard

Site A has abundances (10, 5, 0) and site B (6, 5, 3) of three species. What is the Bray–Curtis dissimilarity?

1. 0.76
2. 0.38
3. 0.24
4. 0.11

9 · Multivariate methods: PCA, cluster analysis, ordination · medium

An NMDS ordination of plot communities has a stress of 0.25. This indicates:

1. An excellent fit that explains 75% of the variation
2. A perfect representation of the dissimilarities
3. That the arch effect has been removed
4. A poor fit; interpret the ordination with caution

10 · Research ethics, data management and reproducible analysis · medium

A researcher runs 10 independent tests and applies a Bonferroni correction for an overall α of 0.05. Each test uses:

1. 0.05
2. 0.5
3. 0.005
4. 0.0005