GGA training session 3 – Trial Protocols and Best Data Practices

Adam H. Sparks

Curtin Biometry and Agricultural Data Analytics

September 5, 2024

On-Farm Research Success

1. Ask the Right Question (refer to Mark’s session this AM)
2. Ask what data will be collected
3. Consider paddock history and variation
4. Pick the right trial design (refer to Point 3 above & previous session)
   1. Include a control treatment for a baseline!
5. Replicate and randomise (refer to previous session)
6. Collect the data
7. Share the raw data with AAGI

Ensuring You Have Quality Data

1. Ensure the trial design is valid and fit for purpose.
2. Ensure the trial design is recorded correctly.
3. Identify and control sources of noise.
4. Implement protocols to avoid mistakes in running the trial and recording of observations.
5. Always provide the raw data to the biometrician.
6. Record ancillary details.

Don’t Optimise for Convenience

Source: Dr Karyn Reeves, SAGI-West

Don’t Optimise for Convenience

Source: Andrew Van Burgel, DPIRD

Ensure the Design is Recorded Correctly

Source: Dr Karyn Reeves, SAGI-West

Ensure the Design is Recorded Correctly

Source: Dr Karyn Reeves, SAGI-West

Controlling Measurement Error and Uncertainty

Source:SVGRepo

Controlling Measurement Error and Noise

Source: Dr Karyn Reeves, SAGI-West

Source: Dr Karyn Reeves, SAGI-West

Controlling Measurement Error and Noise

Source: Dr Karyn Reeves, SAGI-West

Controlling Measurement Error and Noise

Source: Dr Karyn Reeves, SAGI-West

Source: Dr Karyn Reeves, SAGI-West

Always Provide the Full Set of Raw Data

Source: Dr Karyn Reeves, SAGI-West

Record Ancillary Details

Source: Dr Ben Biddulph, DPIRD

Check Your Data

Source: Dr Karyn Reeves, SAGI-West

Effective Use of Spreadsheets for Sharing Data With AAGI

Best Practices to Follow

• All columns supplied with the trial design should be retained

• Every plot needs a unique identifier (e.g., plot number)

• Avoid copy-paste (easy to introduce errors)

• Ensure embedded equations reference the correct columns

  • Or better yet, don’t do calculations in the spreadsheet at all

Source: Dr Karyn Reeves, SAGI-West

Basic Rules for Data in Spreadsheets

• Be Consistent
  • Use the same layout if you have multiple files
  • Use the same variable names (also if you have multiple files)
  • Only use one variable name, e.g., S10 is different than S 10

From Broman and Woo (2018)

Choose Good Names for Things

Good Name	Good Alternative	Avoid
max_temp_C	MaxTemp	Maximum Temp (˚C)
precipitation_mm	precip	precmm
mean_year_growth	MeanYearGrowth	Mean growth/year
yield_kg_ha	yield	yield kg/ha
observation_o1	first_observation	1st Obs.

The good names include the measured variable and the units. The alternatives are less descriptive but are clear and do not have spaces in them. The avoid have spaces and upper and lower case or are ambiguous.

After Broman and Woo (2018)

Write Dates as YYYY-MM-DD (ISO 8601)

A spreadsheet with inconsistent date formats. This spreadsheet does not adhere to recommendations for consistency of date format. From Broman and Woo (2018). Also see https://xkcd.com/1179.

No Empty Cells

Examples of spreadsheets that violate the no empty cells recommendation. (a) A spreadsheet where only the first of several repeated values was included. (b) A spreadsheet with a complicated layout and some implicit column headers, from Broman and Woo (2018).

Put Just One Thing in a Cell

• e.g., Don’t use rep-plot for a header and use 1-1, 1-2…2-1, in the column, etc.
• Do use rep and plot as headers and have the values in separated columns

Make It a Rectangle

Examples of spreadsheets with nonrectangular layouts. These layouts are likely to cause problems in analysis, from Broman and Woo (2018).

Example A, what not to do.

Keep your data tidy and in a rectangle so that it’s easy to analyse when you send it to AAGI. These examples have many empty cells in-between values. Computers don’t like these sorts of holes in the data (and thus neither do the biometricians using the computers).

Make It a Rectangle

A spreadsheet with two header rows. It is better to have a single header row, from Broman and Woo (2018).

Example B, what not to do.

This is a common example of having two header rows. Following is an example of how to integrate this header row into a column for a nice rectangular data set.

Make It a Rectangle

An example spreadsheet of the previous example’s data in a rectangular layout, from Broman and Woo (2018).

Yes, do this!

We can see here that we’ve incorporated the week heading into a column, yes, it’s repeated but that’s what the computer will look for when the analysis is run and will not affect anything. In fact, it’s preferred!

Create a Data Dictionary

An example data dictionary, from Broman and Woo (2018).

My suggestion is that this goes in the trial protocol that is provided to AAGI with the data.

Do Not Use Colour as Data

Highlighting in spreadsheets.(a) A potential outlier indicated by highlighting the cell.(b) The preferred method for indicating outliers, via an additional column, from Broman and Woo (2018).

Save Data in Plain Text Files (e.g., CSV)

(a) An example spreadsheet. (b) The same data as a plain text file in CSV format, from Broman and Woo (2018).

Lastly

• No calculations in the raw data file
• Make backups
• Use data validation to avoid errors

Avoid calculations in the raw data that you send to AAGI. Just provide the raw data, AAGI will perform the calcuations to create mean values and other summaries as necessary. Performing calculations in the spreadsheet can lead to errors and obscure the raw data if it’s not provided to AAGI pre-calculated.

Make backups!

Excel can validate your data. I won’t go into it here, but it has the capabilities to enforce cell types to help you ensure that you don’t put a date in a yield column that should be a number not a date, or a weed name (text) in a weed count column (numeric).

From Broman and Woo (2018)

Exercise

Exercise (20 min)

Working with your partner on the experiment you designed this morning.

1. Create a trial protocol that details the:
   1. Treatments and
   2. Data to be collected.
2. Draw or list the spreadsheet headings for the data you will be collecting to send to AAGI.
   1. Describe the date format you will use,
   2. Describe what you will use to describe missing data and
   3. Describe what descriptions need to be included in the data dictionary/protocol for AAGI to refer to.

Thank You

References

Broman, Karl W., and Kara H. Woo. 2018. “Data Organization in Spreadsheets.” The American Statistician 72 (1): 2–10. https://doi.org/10.1080/00031305.2017.1375989.