What Causes Errors?


When we see inconsistencies in data collected from the same field, or in the same season, we rightly ask why the data are not more precise or more accurate. Or to put it another way, what can we do to improve the quality of the data we are collecting?

Going back to the example of yield data from Lakeland College we discussed above, in an ideal world if we knew the true yield value, then we could know how “off” the measured yield is from the true value and therefore, how accurate the data we are collecting actually is. Unfortunately, this is not easily done in practice.

We can break errors into two categories: 

1. Random error: A random error is a chance difference between the observed and true value (e.g. a researcher misreading a weighing scale).

2. Systematic error:  A systematic error is a consistent difference between the observed and true value (e.g. a miscalibrated scale consistently registers a higher weight). 

Now let’s take a closer look at how these errors may emerge by reviewing some examples.  Keep in mind that this set of examples is not exhaustive. There are many ways that both random and systematic errors can occur in your data collection practices.


Random Errors

Natural variation in context: During a canola trial the test plots are exposed to variable conditions due to their shape, orientation and proximity to a forested area. This natural variation results in some random error in plant growth and yield. 

Imprecise Instrument: The capabilities of the sensor or instrument leads to a certain amount of inherent uncertainty and error in the data produced. For example, the accuracy of different GPS technologies can vary significantly, running from 100 meters to less than a meter.

Individual differences: Observations, especially qualitative observations {see  📓⁠⁠Categorizing Data⁠  in Module 1), made by people are always subject to a certain amount of subjectivity. For example, if you conduct a pasture quality test on your farm the results you obtain will vary slightly from the results produced by another person, even if you both make your observations in the same pasture on the same day.


Systematic Errors

Calibration Error:  A poorly calibrated sensor or instrument will give consistently inaccurate readings.  For example, a scale that is not zeroed or tared properly will give a consistently inaccurate reading of the mass, or weight, on the scale.

Experimenter drift: Qualitative data collected by people is vulnerable to drift as the data collector becomes fatigued or distracted. For example, field scouting is labour intensive and requires a great deal of focus. Over the course of a day, the data collector can lose focus and may become less diligent with data collection protocols, resulting in drift in the data being produced.

Sampling bias:  It is not usually possible to measure the entire population of interest, as a result you may choose instead to measure a sample from the larger population. Unfortunately, the process of selecting the sample can introduce some error into the process. In addition to any potential random error associated with choosing the sample, a sample bias can occur if the sample is chosen in a way that isn’t representative of the whole. For example, extracting a grain sample from the top of the pile in the grain truck to perform quality analysis because it is convenient may create a sampling bias if it isn’t representative of the grain in the larger shipment being transported by the truck.


A certain amount of error is unavoidable in measured data - the error rate will never be zero.

However, we can take deliberate and strategic actions to limit the impact of errors: 
  • Checking sensors after installation or at the beginning of a data collection operation to ensure data is being transmitted properly.
  • Reviewing data regularly to ensure that it is reasonable and within an expected range. For example, you may question the quality of precipitation data from a weather station when it registers a 100 mm rainfall event over a 1 hr period without a storm passing through the area.
  • Using a secondary instrument to confirm measured values obtained from the primary instrument (e.g. ground truthing data from remote sensing).

Finally, it is important to remember that reducing the errors present in your agriculture data is a learning process. Incremental improvements can have a large impact over time.



Now that we've gone through data accuracy and precision, let's test out your understanding with a short quiz!




Next:  ⚙️⁠⁠Data Lifecycle in Agriculture⁠