Finding Lower-Tail and Two-Sided Z-Test P-Values

Why the Sign of the z Statistic Matters

A one-proportion z test compares a sample proportion with a value specified by the null hypothesis. The z statistic tells us how far the sample result is from that null value, measured in standard errors. Its sign tells us whether the sample proportion is above or below the null value. To calculate a p-value, we also need the alternative hypothesis: it identifies which results count as evidence against the null.

For a lower-tail alternative, small values of the statistic are evidence against the null, so the p-value is the standard normal area to the left of the observed z statistic. For a two-sided alternative, departures in either direction count, so the p-value includes extreme values on both sides. These rules depend on the direction of the alternative, not on which direction happens to look more convenient after seeing the data.

The One-Proportion z Model

Before using a z-test p-value, confirm that the normal model for the test statistic is reasonable. Suppose each observation is a binary outcome, with “success” defined in advance, and the null hypothesis specifies population proportion \(p_0\). The sample proportion is \(\hat{p}=x/n\), where \(x\) is the number of successes among \(n\) observations. Under the null, its standard error is calculated using \(p_0\), not \(\hat{p}\).

The p-value procedure uses the approximation \(Z\sim N(0,1)\) for the test statistic under \(H_0\). This approximation is appropriate when the data come from a suitable random sample or randomized process, observations are independent (or the sampling design justifies treating them as approximately independent), and the null expected numbers of successes and failures are sufficiently large. A common check is \(np_0\ge 10\) and \(n(1-p_0)\ge 10\). These checks concern the null model; a large observed count alone does not establish that the approximation is suitable.

The standard normal cumulative distribution function, often written \(\Phi(z)\), gives the area to the left of \(z\): \(\Phi(z)=P(Z\le z)\). A table or calculator may report this left-tail area directly. If it reports left-tail areas, an upper-tail area is \(P(Z\ge z)=1-\Phi(z)\). For a continuous distribution, using \(<\) instead of \(\le\) does not change these probabilities.

Lower-Tail p-Values

For a lower-tail test, the alternative is \(H_a:p<p_0\). Values of \(\hat{p}\) below \(p_0\) give negative z statistics, and more negative statistics are increasingly far in the direction specified by the alternative. The p-value is the probability of getting a statistic at least that far into the lower tail if \(H_0\) is true.

Two-Sided p-Values

For a two-sided test, the alternative is \(H_a:p\ne p_0\). A sample proportion much smaller than \(p_0\) and one much larger than \(p_0\) both count as departures from the null. The two-sided p-value adds the tail area at least as extreme as the observation on each side of zero.

The two equivalent tail expressions follow from the symmetry of the standard normal distribution. In terms of the observed sign, if \(z>0\), double the upper-tail area beyond \(z\); if \(z<0\), double the lower-tail area below \(z\). Thus, a positive observed statistic does not turn a two-sided test into an upper-tail test: it merely tells us which tail is the smaller one to double.

Why the Two-Sided Rule Works

The two-sided formula is a consequence of symmetry and the definition of “at least as far from zero.” Let \(a=|z|\), so \(a\ge 0\). A standard normal statistic is at least as far from zero as the observed statistic exactly when it is no greater than \(-a\) or no less than \(a\). These events do not overlap, so their probabilities add.

The standard normal density is symmetric about zero, so reflecting values on the right of \(a\) to the left of \(-a\) preserves area: \(P(Z\ge a)=P(Z\le -a)\). Therefore,

If the observed \(z\) is negative, then \(-a=z\), giving \(2P(Z\le z)\). If it is positive, then \(a=z\), giving \(2P(Z\ge z)\). This also explains why the absolute value is essential: the two-sided probability depends on distance from zero, not on the sign alone. At \(z=0\), the probability is 1, since every possible value of \(Z\) is at least zero units from zero.

A Reliable Calculation Routine

The following sequence keeps the hypothesis direction, the observed sign, and the normal tail area aligned. Retain the unrounded statistic during the calculation when possible, and round the reported p-value to an appropriate number of significant figures.

Common Mistakes and Why the P-Value Matters

A frequent mistake is to use the tail corresponding to the sign of \(z\) without first checking the alternative. A negative \(z\) does not automatically call for a lower-tail p-value: it does for a lower-tail alternative, but a two-sided alternative requires including both extreme tails. Similarly, a positive \(z\) in a two-sided test requires twice the upper-tail area, not just the upper-tail area.

Another error is to double the larger tail. For \(z=1.60\), for example, the upper-tail area is about 0.0548, whereas the lower-tail area is about 0.9452. Doubling the larger area would give 1.8904, which cannot be a probability. The two-sided p-value must be the probability in the two tails beyond \(-|z|\) and \(|z|\), so it cannot exceed 1.

A p-value summarizes how unusual the observed statistic, or one more extreme under the alternative, would be if the null model were correct and the assumptions held. It does not measure the size or practical importance of a difference in proportions, and it is not the probability that the null hypothesis is true. Correctly matching the tail to the alternative is essential: even perfectly read normal probabilities answer the wrong question if the wrong tail is chosen.