What it means
Statistical tests start with a null hypothesis, which is the cautious default that nothing has changed. The analyst then collects a sample and asks how likely the result would be if the default were true.
If that probability, called the p-value, is below a chosen level such as 5%, the default is rejected. In a one-tailed test, the alternative claim points in a single direction.
A marketing manager might test whether a new checkout page raises average order value, and a quality controller might test whether a defect rate has fallen. The 5% risk is concentrated in one end of the distribution, so the evidence needed to reject the default is lower than in a test that looks both ways.
The direction must be chosen before looking at the data. If an analyst waits to see which way the numbers move and then picks the matching tail, the test is biased and the real error rate is higher than reported.
This is a common way for results to look more convincing than they truly are. A two-tailed test asks whether the result is different in either direction, and it is the safer default when a move either way would matter.
Use a one-tailed test only when a change in the opposite direction would lead to the same action as no change at all. For example, if a cheaper supplier is acceptable as long as quality is not worse, only a drop in quality needs testing.
For finance teams, these tests appear in forecasting checks, audit sampling, A/B tests on pricing and reviews of fund performance. They give a disciplined way to decide whether a difference is real or simply the result of random variation.
The answer should always be read together with the size of the effect and its cost.
In practice
Real-world examples.
Example
An online retailer changes its checkout page and wants to know if average order value has risen above $50. It uses a one-tailed test because a fall would simply lead it to revert the change. The test shows a significant increase, and the page is kept.
Example
A manufacturer tests whether a new process reduces the defect rate below the current 4%. After inspecting 2,000 items, the defect rate is 3.4%, and the one-tailed test indicates the improvement is unlikely to be luck. The company rolls the process out across its plants.
Example
An auditor tests whether the error rate in a sample of 400 invoices exceeds the 2% tolerance level. Only errors above the tolerance matter, so she uses a one-tailed test. The result is not significant, and she records no further action.
Formula
Calculation
Z = (sample mean - hypothesised mean) / (standard deviation / square root of sample size)
Reject the default (at the 5% level, one-tailed, upper tail) if z is greater than 1.645.
Suppose the average order value has been $50. After a website change, a sample of 100 orders has an average of $53, and the standard deviation of order values is $15.
Standard error = 15 / square root of 100 = 15 / 10 = $1.50.
z = (53 - 50) / 1.50 = 3 / 1.50 = 2.0.
Since 2.0 is above 1.645, the result is significant and the default is rejected, with a p-value of about 2.3%. A two-tailed test at 5% would require z above 1.96, which 2.0 also passes, but only just.Case study
Seen in the real world.
Pinewood Lending is an illustrative, fictional lender that trialled a new credit score to see if it would reduce loan defaults below the existing rate of 6%. The risk team tested 5,000 loans approved under the new score.
The default rate in the trial was 5.4%. The team stated in advance that they would only adopt the score if defaults fell, so they used a one-tailed test at the 5% level, and the result was significant.
The chief risk officer was cautious, however. She pointed out that a 0.6 percentage point improvement on a loan book of $200 million is worth about $1.2 million a year, and that the new score costs $400,000 to run. The score was adopted, but a second review was set for twelve months later. The illustrative lesson is that statistical significance is only the start of the business decision.
Watch out
Common mistakes.
- Choosing the direction of the test after seeing which way the data moved, which makes the result look stronger than it is.
- Using a one-tailed test to rescue a result that just failed a two-tailed test.
- Treating statistical significance as proof of a big or valuable effect, when a tiny difference can be significant in a large sample.
Questions
People also ask.
When should I use a one-tailed test?
Use it when you only care about a change in one direction and a change in the other direction would lead to the same decision as no change.
What is the difference in the critical value?
At the 5% level, the one-tailed critical z value is 1.645, while the two-tailed value is 1.96.
What does the p-value tell me?
It is the probability of seeing a result at least as extreme as the one observed if the default assumption were true, and a small value is evidence against that assumption.
From the founder's library
Take it further with the book.
Build your financial confidence beyond this definition. Shihan's full-length guide, Accounting Fundamentals, takes the same plain-English approach and turns it into a complete, practical playbook for non-finance managers, business owners and students - with chapter-end quiz answers and presentation slides included.
25% off with code MMHQ25, applied at checkout. Priced in USD - checkout may show the equivalent in your local currency.
View the book and save 25%Related