Consulting Service at the Institute

We highly recommend to arrange a first consultancy date before you set up your study or experiment to discuss the statistical aspects of experimental design. The prerequisite of any statistical evaluation is the randomized replication of treatments of interest in experimental settings or a proper plan for obtaining random samples in observational studies

When you describe your experiment, think of the following five issues:

1. Experimental units: What is your experimental unit? In a lab experiment, the experimental units might be e.g., single plants, animals, or cell cultures in petri dishes. In a field or greenhouse experiment, they might be e.g., single plants, plots, or entire fields.

2. Randomization structure: Randomization means that the treatments of interest are repeatedly and randomly assigned to the single experimental units, at least at some level of the experiment. Name your randomization structure unmistakably (“completely randomized”, “split-plot”, “block design”, …) or describe it concisely (a picture or drawing might help you). How have experimental units been arranged in time and space? What sub-structures are involved in that arrangement (e.g., technical replications, sub-sampling, …)?

3. Response variables: Which variables measure the effect you are interested in? Are they measured in continuous numbers, counts of events or individuals, percentages/proportions, ratios of two primary variables, …?

4. Explanatory variables: Which variables or factors are supposed to explain differences in the response variables? How many factor levels are there? Which factors/variables are of primary interest?

5. Experimental question: Explain the purpose of your research as precisely as possible. If you're not quite sure about the experimental question of interest, talk to your supervisor first! “I want to find out if increasing N fertilization has an impact on the leaf area”, is a sensible research goal. “I want to find significant p-values”, is not.

• If possible, submit your data as txt or csv file. Excel-type xls files can be accepted as well.
• Data that should be analyzed together should (normally) be in one single file (if you provide an xls file: in one single sheet; exception, e.g., high dimensional data).
• Record your data in flat file format: there should be one separate line for each experimental unit (plant, animal, pot, petri dish, …) in your data table i.e., all information regarding a certain experimental unit should be in one row. The columns should contain the values of observed variables and controlled factors as well as information in which partial experiment or randomization structure (e.g., in which block) the particular experimental units were placed (exception: highdimensional data).
• All columns should have (short) headers. They may contain letters and numbers but no special characters (ä ß µ % $), mathematical operators (+ - * / ^), brackets, or blanks. Underscores are okay.
• Code missing values by “NA” and not by numerical values like 0 or 999.
• Denote all structures in the experiment which might influence the outcome by introducing a new variable (i.e., a new column), even if these variables are not of direct interest (e.g., if parts of the experiment were performed in different weeks, introduce a new column that codes which observation comes from which week).
• If you submit an xls file, your data sheet for statistical analysis should not contain:
  • formulas, calculations or graphs
  • comments and explanations
  • blank lines, empty columns or empty cells within the table
  • colors, fonts, boldface, etc. as the only distinction of treatments or replications
  • links to other documents
  • several tables in one sheet

[Translate to English:] Ein Datensatz im flat file format, bei dem alles eindeutig deklariert ist, könnte so aussehen: