Previous Next

Plant Detectives Manual: a research-led approach for teaching plant science

Make the most of your research efforts!

(A statistician and former Plant Detective’s perspective)


Before you embark on this Plant Detectives Project it is worth taking a few minutes to think about some key principles in experimental design and analysis. The project you are about to do has ample opportunity for independence and creative thinking. To make the most of your efforts it is important to give some thought to how you design your experiments and collect your data. Along the way you will also learn about how to analyse your data. The quality of your results, however, flows directly from the care that you have taken in experimental design and data collection.


Statistics is a philosophy as well as a quantitative tool. Modern statistical thinking in laboratory experiments is less than 100 years old. Much of the seminal work in developing the philosophy of modern statistics is credited to R.A. Fisher, also a leading figure in the field of genetics. Fisher was based in Rothamsted Experimental Station from 1919 to 1933, a research institute that continues to be a hotbed of agricultural research. It was during this time that he codified principles of experimental design and analysis. Unfortunately his writing is somewhat turgid and while his seminal work (Statistical Principles for Research Workers (1925)) is a classic, it is not read as widely as it should be.


There are some important basic principles in experimental design and analysis that are central to statistical thinking. These principles follow from the understanding that many uncontrolled factors influence the outcome of experiments. These factors include, for example, the effects of varying temperatures, humidity, greenhouse positioning, soil type and variation among individual plants. Following the basic principles can help researchers to design more efficient experiments and eliminate systematic biases that may produce misleading results.

  1. Controls direct comparison against a known standard. A control may consist of absence of treatment, sham treatment or placebo. Control should be concurrent and tested under identical conditions to treatment. In this set of experiments, the controls are the wild type plants (or leaves or seeds thereof).


Whenever you set up an experiment, ask yourself: How can I ensure that wild type and mutant seedlings (in pots or germination plates) are exposed to identical conditions.


  1. Blocking means evaluating genotypes/treatments under homogeneous (low variability) experimental conditions. Treatment effects are measured within blocks and averaged across blocks. A block can be a set of units that share characteristics that may influence outcomes. Examples include a tray of seedlings, a technician responsible for processing a set of units, or units that are processed in a batch or run. Your data has ‘structure’ and your plants have a history, like a character in a novel has a back story — blocking can help you to keep a record of that history (e.g., where you place them, how you arrange your genotypes and treatments, who measures what) so that you have the option to account for it in your analyses. Utilising blocking can improve precision and lead to more efficient experiments.


For example:

  1. When you begin your germination assay, determine whether you have a blocked design.
  2. When arranging your plants in trays, ask yourself: How can plants in this tray be arranged to avoid a systematic bias in outcomes?
  3. You are working in a group of other researchers — this is another source of bias. Discuss amongst yourselves how you can best avoid a systematic bias in outcomes?


  1. Replication relates to the number of independent units within a treatment/genotype. The more independent units in the randomised experiment, the more information there is about the treatment, and the more precisely the treatment effect can be estimated. Repeating a measurement on a single unit can also add information about the treatment effect, but this is often not as informative as collecting the same number of measurements in independent units. For example, a leaf collected from a single plant that is split into three samples carries less information than leaf samples from three separate plants. Likewise, technical replicates in a spectrophotometer are not as informative as replicates from separate plants (also called biological replicates).


It is important to distinguish these types of replicates in the statistical analysis. If not, one risks making incorrect inference. In the data analyses of this practical, you will learn how to analyse these data using Analysis of Variance (ANOVA).

  1. When collecting leaf samples for pigment assays, how will you sample leaves to ensure that you obtain three independent samples from each genotype?
  2. In the germination assay, what are the independent units?


  1. Randomisation relates to a probabilistic process in which independent units are assigned to treatments. Randomisation helps to remove biases that may arise from uncontrolled factors, such as position in the tray or on the germination plate. In this practical, there are two situations where randomisation can be employed. Plants can be randomised to positions within a tray. In addition, when half of the plants of each genotype are chosen to undergo drought conditions, a randomisation algorithm rather than purposeful selection can be utilised. When randomisation is used, one can infer that any statistically significant difference between groups can be attributed to the genotype or treatment received.


For example:

  1. What potential biases could arise if mutant seedlings were always on the left side of the tray and wild type seedlings on the right side?
  2. What randomisation method could be used to select which plants will receive drought/normal conditions?


  1. Blinding masks the treatment assignment during assessments. Did you know that it is a well-observed fact that studies that are unblinded report larger treatment differences than studies that are assessment blinded? You can overcome this problem in your experiments.


For example:

  1. What system can you and your colleagues use when collecting the phenotypic data in the seedling experiments? How can you ensure that the evaluator remains ‘blind’ to the genotype of the plant that is being measured?
  2. Similarly in the phenotypic assessments in the germination plate assays, how can the evaluator remain blinded to genotype?


Dr. Terry Neeman

The Australian National University Statistical Consulting Unit

(and former plant detective)


Previous Next