You make decisions every day. Which antibiotic should I use to treat a case of mastitis? Should I add this feed additive? Should we operate on this displaced abomasum or sell the cow?
Most decisions are based on previous experiences and apparent outcomes. In some instances, this may be a reasonable option. In other cases, you may make a significant error.
So, how do you know if your decision is correct — or at least have some probability that your decision is reasonable?
Good or good enough
First ask how confident you need to be in your decision. This is the so-called “p” value used in scientific literature.
Remember that a result may be biologically or clinically relevant, but not statistically significant. If the potential exists for significant economic return for low input, then being wrong carries less risk.
This may come into play with the choice to use one antibiotic treatment over another. If on-farm data support a numerical difference between two treatment protocols, but you are only 80 percent confident in the results, choosing one product over the other may be a reasonable decision because the cost of change is insignificant and there is little risk of failure.
However, be careful in inferring that there is a change when none likely exists. On most dairies, the number of cases needed to make a confident decision specific to your situation is difficult to obtain. For instance, if you want to determine if there is a difference between two synchronization programs, you need about 700 total breedings to make a fair comparison.
Use existing data
Furthermore, just because the last cow treated with antibiotic “X” did not respond does not mean the treatment will be ineffective in the future. You need to have enough cases to make a statistical decision. On-farm and university trials are a good place to start to gain the needed data.
Test results are another area where incorrect decisions occur. All tests are subject to both false-positive and false-negative results.
Use the urine dipstick to detect ketosis as an example. The test is very specific, but has a lower sensitivity. This means the test has a higher level of false-negative results (the test may be negative, but the cow can still have ketosis) but a lower rate of false-positives (the test is likely correct in finding a ketotic cow).
Work with your veterinarian to correctly interpret the results of different tests.
Use the same concept when reviewing production or financial data. Herd data can easily be summarized, but it often takes further calculations and thought to determine if the increase in pregnancy rate or milk production is significant.
Software can help
Confidence intervals are now calculated by some versions of on-farm management software. The confidence interval gives you a range of probable values for the difference between two averages.
Let’s compare the conception rate of two AI technicians. On-farm software may show a numerical difference in conception rate between two technicians, but are you confident enough in these values to terminate one technician?
Confidence intervals give the probable range in conception rate for a given percent certainty (such as 95 percent). If the confidence intervals overlap, you can be quite certain that no conception rate difference exists between the two technicians, even though there is a numerical difference.
On the other hand, if the confidence interval changes, you may be certain that there is a difference and a change in who breeds the cows will be made. Again, biological versus statistical decisions may require individual judgment.
I often see producers make decisions with little supportive data. Seek the expertise of consultants who are familiar with both on-farm and university trials, who can help you make decisions based on science and statistics versus random experiences.
Mark J. Thomas is a veterinarian and partner in Countryside Veterinary Clinic, LLP in Lowville, N.Y.