How to identify outliers to raise herd performance

Targeting average numbers to track performance and monitor changes takes too long to have impact – it is also misleading because a herd’s average can hide a huge range in output.

Farm owners would be better to interrogate their data more thoroughly to identify outliers and work on improving the bottom tier.

“To improve metrics, we need detail to make progress – averages don’t do it,” says vet Dan Humphries of Horizon Dairy Vets in Shropshire.

See also: How to use herd data to create effective change on farm

“An average figure doesn’t tell you where the opportunities for change in your herd are, or what’s really going on.

“Averages also do not reflect fine tuning, where 1% incremental improvements are gained.

“Be aware that averages don’t tell you much of a story,” he says.

Mean averages hide the finer detail

Yet averages are everywhere, routinely used for comparisons and trends – whether talking down the pub or benchmarking in a discussion group.

As a result, Dan thinks there is plenty of lost genetic potential happening in many dairy herds, reflecting that herds don’t have “average” problems, they have “range” problems.

Mostly, the type of average used for making comparisons is an arithmetic mean: calculated by adding all of the values (for example, individual milk yield) then dividing by the total number of values (such as the number of cows). It is this “average” number that “hides things”, says Dan.

To explain this, he shows what is happening behind average milk yield for two large, housed herds (see table below).

They appear to be more or less the same but, on taking a closer look, Herd 2 has a much smaller gap between its top-performing cows and those at the bottom, Dan points out.

This means they are more uniform in output.

“This herd has the same ‘average’ performance, but its feed costs are £1 a cow a day less,” he says.

A wide range in performance is harder to manage

Such huge ranges in performance make farm management difficult, he points out.

“If you have a consistent group of cows, you can match the nutrients closer.

“But a big spread shows that different pressures – such as enough feed space – are not giving all cows in the herd the same opportunity to perform and express their genetic potential,” he explains.

Cows deserve to be compared on an even footing and benefit from management that keeps up with rising genetic potential.

“We compare cow genetics, yet we don’t compare their individual opportunity to eat, visit the parlour, lie in a cubicle,” he adds.

“And we wouldn’t see any of this if we only look at the ‘average’ numbers. The key is to look at opportunities to improve things, and this starts by looking at why there is a massive spread [in results].”

This is why it is important to drill down into the specifics of a farm’s data to discover the reasons for different output from similar genetics.

Some cows may simply lack feeding space, others may have had disease as a calf, which led to poor growth rates and, subsequently, substandard performance compared with their peers.

How to identify outliers in the herd

Using a farm’s data to show a standard distribution bell curve (where results are based around the average in the middle), Dan says it is possible to pick out the outliers.

These are the cows at the top and bottom of the herd, or the “above average” and “below average” performers.

The farm team should then specify what success looks like and the percentage of animals to hit that target.

By selecting the right proportion of cows to target, it means that management is not trying to improve the top 25% of cows, nor cull out the bottom 25%, he says.

The bottom cohort will simply be replaced by other cows if management fails to change, leaving “the root cause of the problem” still to fix.

“They might need to go, but address why they are there in the first place,” he advises.

Timeframes for success

Some change takes time, however, and Dan uses average age at first calving as an example.

If the goal is to calve replacements at 24 months, it may be that heifers are poorly grown at 13 months of age.

This will not be rectified overnight, so there is a lag time before a difference can be seen.

On the other hand, if heifers are well grown and they are not getting in-calf because of slack heat detection, this situation can be fixed overnight with wearable technology.

Targeting averages takes time for alterations to catch up and make a difference, whereas Dan says it is important to be on the ball and respond to setbacks as they arise.

Therefore, changes and fixes should be tailored to the individual farm according to what they wish to achieve.

He advises farmers to look for progress in their own system, not to just achieve a KPI then target the next one.

“Industry KPIs are interesting but not detailed enough to help farmers identify areas for change. Some are good for comparing farm to farm, as long as everyone is measuring the same way.”