Why a Crossover Point Matters
When two groups have different fitted slopes, the fitted difference between them can change as the predictor changes. A crossover point is the predictor value at which the two group-specific fitted lines give the same fitted outcome. It marks where their ordering switches, if the lines actually meet and then separate.
The algebra for locating that point is simple. The important judgment is whether it is relevant to the data. A crossing can fall within the predictor values observed for both groups, outside the observed values, or in a gap where neither group has observations. Only the first situation gives direct data support for comparing the groups at the crossing.
Definition and Formula
Consider two groups, \(A\) and \(B\), and a quantitative predictor \(x\). Suppose their fitted outcomes are represented by straight lines on the same outcome scale:
Here, \(\hat{a}_A\) and \(\hat{a}_B\) are the fitted intercepts, and \(\hat{b}_A\) and \(\hat{b}_B\) are the fitted slopes. The fitted group difference, defined here as group \(B\) minus group \(A\), is:
The formula follows by setting the two fitted outcomes equal and collecting the terms involving \(x\):
The denominator is the difference between the fitted slopes. If it is zero, the lines are parallel: they never meet when their intercepts differ, and coincide at every predictor value when their intercepts are also equal. There is then no unique crossover point.
Once \(x^*\) is found, the common fitted outcome is obtained by substituting \(x^*\) into either line. The fitted difference can also be written as \(\hat{D}(x)=(\hat{b}_B-\hat{b}_A)(x-x^*)\) when a unique crossing exists. This expression makes the change in ordering clear: the sign of the fitted difference reverses at \(x^*\).
Worked Examples
Worked Example: A Crossing Within Both Groups’ Ranges
Suppose a hypothetical analysis relates hours of practice to a task score, measured in points, for two training groups. The fitted lines are \(\hat{y}_A(x)=52+4x\) and \(\hat{y}_B(x)=70+1x\), where \(x\) is practice time in hours. The observed practice-time range is 2 to 10 hours in group A and 4 to 9 hours in group B. Find and interpret the crossover.
Set the fitted lines equal:
Substituting \(x=6\) into each line gives \(52+4(6)=52+24=76\) points for group A and \(70+1(6)=70+6=76\) points for group B. Thus the fitted lines meet at 6 practice hours and a fitted task score of 76 points.
Group A’s observed range is 2–10 hours and group B’s is 4–9 hours, so both groups have observations across the common range 4–9 hours. The crossing at 6 hours is inside that shared range. The fitted difference, group B minus group A, is \(18-3x\). It is positive below 6 hours and negative above 6 hours: the fitted score is higher for group B below the crossing and for group A above it, within the predictor values represented by both groups.
Worked Example: A Crossing Beyond the Observed Values
In a second hypothetical setting, two groups have fitted outcomes, measured in units of an index, given by \(\hat{y}_A(x)=30+2x\) and \(\hat{y}_B(x)=18+3x\). Group A’s observed predictor values range from 1 to 8, and group B’s from 2 to 10. Find the crossing and assess its support.
Equating the lines gives
At this value, both fitted outcomes equal \(30+2(12)=54\) index units; the other line gives \(18+3(12)=54\) as well. But 12 is greater than the largest observed predictor value in either group: group A ends at 8 and group B at 10. The crossing is outside both groups’ observed ranges.
The fitted difference is \(\hat{y}_B(x)-\hat{y}_A(x)=-12+x\). For every \(x\) in the overall observed span, 1 to 10, this difference is negative, so the fitted outcome for group A is higher throughout that span. The calculation predicts a reversal at 12 only by extending both lines beyond their observed values. It is an algebraic intersection, not evidence that an observed group comparison reverses.
Worked Example: A Crossing in a Gap Between Group Ranges
Suppose hypothetical fitted lines for an outcome measured in units are \(\hat{y}_A(x)=20+3x\) and \(\hat{y}_B(x)=50-3x\). Group A has observed predictor values from 0 to 4, while group B has values from 6 to 10. Does a crossing occur within the data support?
Solving for equality:
At \(x=5\), both lines give \(20+3(5)=35\) units and \(50-3(5)=35\) units. The crossing lies inside the combined span from 0 to 10, but not in either group’s observed range: group A’s values stop at 4 and group B’s begin at 6. In particular, there is no shared range of observed predictor values for these groups.
Thus the fitted lines meet in a gap between the groups’ data. The crossing requires a short extrapolation beyond each group’s observed values, and there are no observations from both groups near \(x=5\) to support a direct comparison there. Saying the groups’ outcomes are observed to be equal at \(x=5\) would be incorrect; only the two extended fitted lines have equal predictions there.
Worked Example: Parallel Lines Have No Unique Crossing
For a final hypothetical example, let \(\hat{y}_A(x)=40+2.5x\) and \(\hat{y}_B(x)=55+2.5x\). Equating the lines would require
This is impossible, so there is no crossing. Both groups have the same fitted slope, and the fitted difference is \(\hat{y}_B(x)-\hat{y}_A(x)=15\) outcome units at every \(x\). If the two lines instead had the same intercept as well as the same slope, they would coincide and every predictor value would give equal fitted outcomes; there would not be one unique crossing.
Checking Whether a Crossing Is Supported
For each group, the observed predictor range runs from that group’s smallest to largest observed \(x\). The common range is the interval shared by the two groups’ observed ranges, if they overlap. A crossing inside the common range is generally more directly supported than one outside it because both groups have observations within that interval.
Being inside both groups’ minimum-to-maximum ranges is a useful check, but it does not guarantee that observations are plentiful near the crossing. A group might have very few observations there, or observations might be concentrated far from \(x^*\). A plot of the observed values by group can help reveal such gaps. The fitted lines also assume that a straight-line mean pattern is reasonable over the values being interpreted. A calculated crossing does not make that assumption true.
Use the same predictor \(x\), outcome scale, and units for groups A and B.
If the fitted slopes are equal, the lines are parallel or coincident; there is no unique crossing.
For unequal slopes, calculate \(x^*=(\hat{a}_A-\hat{a}_B)/(\hat{b}_B-\hat{b}_A)\), then substitute it into either line for the common fitted outcome.
Compare \(x^*\) with both observed predictor ranges, and look for adequate observations near it, especially within the common range.
Give the predictor value and units, the common fitted outcome and units, and whether the crossing is supported by the observed ranges.
The calculations use estimated fitted lines, so the crossing is itself an estimate based on the data and model. Finding a crossing does not establish that the underlying group mean outcomes truly switch order at exactly that predictor value. The uncertainty in fitted predictions and group differences is a separate consideration.
Common Mistake and Why the Range Check Matters
A common mistake is to calculate \(x^*\), notice that it falls between the smallest and largest predictor values in the pooled data, and conclude that the groups’ fitted outcomes cross where both groups are represented. That conclusion can fail when the groups have different ranges. In the gap example, \(x=5\) was inside the pooled span of 0 to 10, but no group had observations at 5 and the groups had no common observed range.
The correction is to compare the crossing with each group’s range, not just the combined range. Then check whether there are observations near the crossing. A value within both groups’ ranges is more informative for a group comparison than a value supported only by extending the fitted lines; a value outside one or both ranges requires particular caution.
Check Your Understanding
Use the fitted lines and observed ranges in each question. State the units and distinguish a line intersection from a crossing supported by both groups’ data.
- For \(\hat{y}_A(x)=12+5x\) and \(\hat{y}_B(x)=36+2x\), calculate the crossover predictor value and common fitted outcome.
- For \(\hat{y}_A(x)=8+3x\) and \(\hat{y}_B(x)=20+x\), which group has the higher fitted outcome at \(x=4\)? Show the fitted difference.
- Two groups have observed predictor ranges 1–7 and 5–9. Is a crossing at \(x=6\) inside their common range? What additional feature of the observed values would you check?
- What can be concluded about a unique crossover when the two fitted slopes are equal but the fitted intercepts differ?
- Why is a crossing inside the combined range, but outside one group’s observed range, not strong direct evidence of a group reversal?