Influential points and outliers: regression worksheet

8 problems on influential points and outliers: regression, ready to print for class or homework. The answer key prints on its own page, so hand out everything before it.

Problem 1

An urban beekeeper fits a least-squares line to 25 hives, relating the frames of brood counted in spring, xx, to the honey harvested that summer in kilograms, yy. All 25 hives sit close to the line y^=2+3x\hat{y} = 2 + 3x, their frame counts run from 4 to 14, and the mean is xˉ=9\bar{x} = 9 frames.

Treat each hive below as a separate 26th hive added to those same 25. Classify each one as an outlier in the y direction, a high-leverage point, both, or neither.

(a) Hive A: 9 frames, 45 kg. (b) Hive B: 24 frames, 74 kg. (c) Hive C: 26 frames, 24 kg. (d) Hive D: 6 frames, 22 kg.

Problem 2

A vineyard fits one least-squares line predicting a vine's grape yield in kilograms, yy, from the vine's age in years, xx, using all 40 vines in a block. Across those 40 vines the mean age is xˉ=12\bar{x} = 12 years with standard deviation sx=4s_x = 4 years. Every vine that is not listed in the table below has an age between 6 and 18 years and a residual between -4 and +4 kg.

VineAge (years)Residual (kg)
A139.2
B26-0.4
C258.8
D11-1.1

(a) For each vine, find how many standard deviations its age sits from xˉ\bar{x}. (b) Classify each vine as an outlier in the y direction, high leverage, both, or neither. (c) Which vine is most likely influential, and why does vine B's tiny residual not settle the question?

Problem 3

A pick-your-own strawberry farm records, for five weeks of one June, the week's rainfall in centimeters, xx, and the kilograms picked that week, yy: (2,7)(2, 7), (4,13)(4, 13), (6,13)(6, 13), (8,19)(8, 19), (20,13)(20, 13). The 20 cm week was a flood that washed out part of the field.

(a) Find the least-squares slope using all five weeks. (b) Find the slope using only the first four weeks. (c) Compare the two, and say what kind of point the flood week is.

Problem 4

A school district puts solar panels on five storage sheds and records the number of panels on each shed, xx, and one sunny day's output in kilowatt-hours, yy: (1,2)(1, 2), (2,6)(2, 6), (3,6)(3, 6), (4,10)(4, 10), (10,24)(10, 24).

(a) Find the least-squares line using all five sheds. (b) Find the least-squares line again without the 10-panel shed. (c) Find r2r^2 both ways. (d) Is the 10-panel shed a high-leverage point, and is it influential?

Problem 5

A ferry operator plots, for 30 winter crossings, the wind speed in knots at departure, xx, against the fuel burned in liters, yy. Twenty-nine of the crossings had wind between 5 and 22 knots and show a moderate positive linear association. The remaining crossing had 44 knots of wind and burned far less fuel than the pattern of the others predicts, because the captain shortened the route that day.

(a) Classify the 44-knot crossing. (b) If it is removed, does the slope get larger or smaller? (c) What happens to rr? (d) What happens to r2r^2? (e) Suppose instead that the 30th crossing had 40 knots of wind and burned almost exactly what the pattern predicts, with the other 29 crossings unchanged. What happens to the slope and to rr if that crossing is removed?

Problem 6

A survey team fits a least-squares line predicting the number of nesting pairs of terns on an island, yy, from the island's area in hectares, xx, for 18 islands. Using all 18 islands the correlation is r=0.48r = 0.48. One island covers 300 hectares while no other tops 40 hectares, and its nesting count sits well below the pattern of the rest. Refit without that island and the correlation is r=0.86r = 0.86.

(a) Find r2r^2 both ways and interpret each in context. (b) Is the large island influential? (c) Once it is removed, will the slope be larger or smaller? (d) Does a change in rr this big prove the island's data are wrong?

Problem 7

A neighborhood repair cafe logs, for five Saturday sessions, the number of volunteer fixers on duty, xx, and the number of items repaired, yy: (1,4)(1, 4), (2,6)(2, 6), (3,10)(3, 10), (4,12)(4, 12), (10,3)(10, 3). The 10-fixer session was the launch open house, where the fixers spent most of the day training newcomers instead of repairing.

(a) Find the least-squares line using all five sessions. (b) Find the residual for the launch session and the residual for the 4-fixer session, and say which is larger in size. (c) Refit the line without the launch session. (d) Explain how a point can be the most influential one in the data and still not have the largest residual.

Problem 8

A student fits a line predicting a smoothie shop's daily sales in dollars, yy, from the day's high temperature in degrees Fahrenheit, xx, over 30 days. With all 30 days r2=0.41r^2 = 0.41; leaving out one day gives r2=0.88r^2 = 0.88. That day is logged at 104 degrees while every other day falls between 58 and 86 degrees, and its sales sit far below the pattern.

(a) The student wants to delete the day because the fit is better without it. Is that a good enough reason? (b) The student checks the log and finds the shop's thermometer was swapped that morning and the technician wrote 104 where the reading was 74. What should the student do? (c) Suppose instead the reading is confirmed correct, but the shop closed at noon for a power outage. What should the student do? (d) Suppose the log turns up nothing unusual at all. What should the student do?

Influential points and outliers: regression: answer key

  1. (a) Outlier in the y direction only. (b) High leverage only. (c) Both. (d) Neither. Hive C is the one most likely to change the line.

  2. (a) A: 0.25 SD, B: 3.5 SD, C: 3.25 SD, D: -0.25 SD. (b) A outlier only, B high leverage only, C both, D neither. (c) Vine C; vine B's small residual proves nothing because a high-leverage point pulls the line toward itself, so you have to refit without it.

  3. (a) b=0.18b = 0.18 kg per cm. (b) b=1.8b = 1.8 kg per cm. (c) Removing the flood week multiplies the slope by 10, so it is highly influential and high leverage, even though its residual of -2.16 kg is smaller than the +6.00 kg residual of an ordinary week.

  4. (a) y^=2.4x\hat{y} = 2.4x. (b) y^=2.4x\hat{y} = 2.4x, exactly the same line. (c) r2=0.989r^2 = 0.989 with all five sheds and r2=0.900r^2 = 0.900 without. (d) It is high leverage, and the verdict splits: it is influential on the correlation, lifting r2r^2 from 0.900 to 0.989, but it is not influential on the line, which stays y^=2.4x\hat{y} = 2.4x either way.

  5. (a) Both a high-leverage point and an outlier in the y direction. (b) The slope gets larger (steeper positive). (c) rr increases toward 1. (d) r2r^2 increases. (e) In that separate scenario the slope barely changes, but rr and r2r^2 usually drop, since that point widened the x-range while sitting on the pattern.

  6. (a) r2=0.2304r^2 = 0.2304 with all 18 islands (about 23.0 percent explained) and r2=0.7396r^2 = 0.7396 without (about 74.0 percent). (b) Yes, it is influential and high leverage. (c) Larger. (d) No; a large change in rr shows influence, not error.

  7. (a) y^=8.28−0.32x\hat{y} = 8.28 - 0.32x. (b) The launch session's residual is −2.08-2.08 items and the 4-fixer session's is +5.00+5.00 items, so the ordinary session has the larger residual. (c) y^=1+2.8x\hat{y} = 1 + 2.8x, so the slope flips from −0.32-0.32 to +2.8+2.8. (d) A high-leverage point pulls the line toward itself, so its own residual shrinks while the error it creates lands on the other points.

  8. (a) No; a better fit is never on its own a reason to drop data. (b) Correct the value to 74 degrees, refit, and document the change. (c) You may exclude it, but only by naming the day, stating the power outage as the reason, and reporting both fits. (d) Keep it, flag it as an influential high-leverage point, and report the analysis with and without it.

The key gives final answers only. The full worked solution for each problem is on the matching practice set, and every other sheet is on the worksheets page.

More worksheets from this unit