FS2.1.2 Residuals. The residual sum of squares (RSS)
Section
FS2.1 Linear Regression
Level
AS & A level
Statement
Residuals. The residual sum of squares (RSS)
Guidance
An intuitive use of residuals to check the reasonableness of linear fit and to find possible outliers. Use in refinement of mathematical models. The formula RSS = S_yy − (S_xy)²/S_xx. Derivations are not required.
June 2025 · AS Paper 2 (Further Statistics 2) · Q3(e)2 marks
Mark scheme (whole question; this part is marked with the orange bar)
Examiner's comment
Part (e) was answered very well and most secured both marks here. In part (f) some simply substituted the BMI values into their equation from part (e) and wrongly concluded the B had the greater average speed and so B was the quickest and took the shorter amount of time. Others realised that the residuals should play a part but had the wrong signs when appending these to their equation. A good number though found suitable values for the average speeds but seemed to forget that their values for vA and vB were speeds and not times and therefore picked the incorrect runner.
Key points from the mark scheme and examiner's report
RSS = Svv − Sxv²/Sxx (or Svv(1 − r²)) = 0.52
June 2025 · AS Paper 2 (Further Statistics 2) · Q3(f)3 marks
Mark scheme (whole question; this part is marked with the orange bar)
Examiner's comment
In part (f) some simply substituted the BMI values into their equation from part (e) and wrongly concluded the B had the greater average speed and so B was the quickest and took the shorter amount of time. Others realised that the residuals should play a part but had the wrong signs when appending these to their equation. A good number though found suitable values for the average speeds but seemed to forget that their values for vA and vB were speeds and not times and therefore picked the incorrect runner.
Key points from the mark scheme and examiner's report
actual speed = predicted + residual: vA = 5.2 or 5.3, vB = 5.0
A is faster so A took the shorter time
errors: ignoring residuals, wrong signs, or confusing speed with time
June 2025 · Paper 4B (Further Statistics 2) · Q1(c)1 mark
Mark scheme (whole question; this part is marked with the orange bar)
Examiner's comment
This opening question on the topic of linear regression was a good opening question for candidates with many scoring the majority of the marks. Part (c) required candidates to find the residual sum of squares which was usually found correctly.
Key points from the mark scheme and examiner's report
RSS = Syy − Sxy²/Sxx = 24.5
June 2025 · Paper 4B (Further Statistics 2) · Q1(d)2 marks
Mark scheme (whole question; this part is marked with the orange bar)
Examiner's comment
This opening question on the topic of linear regression was a good opening question for candidates with many scoring the majority of the marks. In part (d), candidates had to find a residual using their equation of the regression line. Most candidates were able to answer this correctly, although on occasion a candidate would subtract the wrong way round.
Key points from the mark scheme and examiner's report
ŷ = 21.9 + 3.76 × 3.26 = 34.16
residual = 36.8 − 34.16 = 2.6 (observed minus predicted)
June 2025 · Paper 4B (Further Statistics 2) · Q1(e)1 mark
Mark scheme (whole question; this part is marked with the orange bar)
Examiner's comment
This opening question on the topic of linear regression was a good opening question for candidates with many scoring the majority of the marks. If candidates were going to drop any marks, it was usually in part (e). Candidates were asked to give a reason that might support the claim that a particular value was an outlier using parts (c) and (d). Some candidates did not attempt any form of calculation suing these parts so the mark was not available, whilst others did not attempt a correct comparison – usually by forgetting to square their residual to compare with the residual sum of squares.
Key points from the mark scheme and examiner's report
2.6² ≈ 7 is about 27% of the RSS (24.5), a large share for one of 15 points
must square the residual and quote the comparison numerically
June 2024 · Paper 4B (Further Statistics 2) · Q1(b)2 marks
Mark scheme (whole question; this part is marked with the orange bar)
Examiner's comment
This opening question on the topic of linear regression brought more challenge than expected In part (b), many candidates lost the accuracy mark by using prematurely rounded values of the coefficients obtained in part (a), and some compounded their error by failing to show full
Key points from the mark scheme and examiner's report
June 2024 · Paper 4B (Further Statistics 2) · Q1(c)2 marks
Mark scheme (whole question; this part is marked with the orange bar)
Examiner's comment
This opening question on the topic of linear regression brought more challenge than expected of candidates made good progress on most of this question, with parts (a) and (c) the most commonly fully correct, there were some who got mixed up with the variables and attempted to find the line h on t.
Key points from the mark scheme and examiner's report
Svv = 64678 − 626²/10 = 25490.4
RSS = Svv − Svw²/Sww = 90.9
June 2024 · Paper 4B (Further Statistics 2) · Q1(d)1 mark
Mark scheme (whole question; this part is marked with the orange bar)
Examiner's comment
This opening question on the topic of linear regression brought more challenge than expected justify, the correct decision in (d).
Key points from the mark scheme and examiner's report
Student A's model, since its RSS (90.9) is much less than 980
must be a comparison, not just 'small RSS'
June 2023 · AS Paper 2 (Further Statistics 2) · Q3(d)1 mark
Mark scheme (whole question; this part is marked with the orange bar)
Examiner's comment
of students using S hv for the gradient. There was some confusion in part (d). Although S vv many knew that the sum of the residuals should always = 0, some thought it had to be ⩽1. In part (e) most simply substituted 1.96 into their equation for the regression line
Key points from the mark scheme and examiner's report
the residuals of a least-squares line always sum to zero
some thought the sum just had to be at most 1
June 2023 · AS Paper 2 (Further Statistics 2) · Q3(e)3 marks
Mark scheme (whole question; this part is marked with the orange bar)
Examiner's comment
⩽1. In part (e) most simply substituted 1.96 into their equation for the regression line and completely ignored the information about Pat’s misread of the residual. Those that did consider this and calculated the actual residual to be 1.23 usually went on to complete this part correctly.
Key points from the mark scheme and examiner's report
true residual = 2.27 − 1.04 = 1.23
v = 105.5(1.96) + 35.0 + 1.23 = 243 km/h
most ignored the residual information
June 2023 · Paper 4B (Further Statistics 2) · Q1(c)1 mark
Mark scheme (whole question; this part is marked with the orange bar)
Examiner's comment
This was a generally welcoming first question as the first six marks were procedural and very straightforward, allowing candidates a gentle introduction to the paper.
Key points from the mark scheme and examiner's report
June 2023 · Paper 4B (Further Statistics 2) · Q1(d)1 mark
Mark scheme (whole question; this part is marked with the orange bar)
Examiner's comment
This was a generally welcoming first question as the first six marks were procedural and very straightforward, allowing candidates a gentle introduction to the paper. part (d), failing to gain the mark. However, this was almost the only error made. The written answer required in part (d) was very high demand. As such, very few candidates answered correctly. The vast majority wrote something general about residuals along the line of most similar questions from previous years, perhaps indicating a lack of understanding of this element of the course.
Key points from the mark scheme and examiner's report
positive residuals at both ends: the model underestimates the 500 m times of the fastest and slowest children
most wrote generic comments about residual patterns
very few correct
June 2022 · AS Paper 2 (Further Statistics 2) · Q3(d)2 marks
Mark scheme (whole question; this part is marked with the orange bar)
Examiner's comment
In part (d), most realised that they needed to substitute x = 32 into their regression equation, but fewer realised the needed to add 80 to find the weight of the bream.
Key points from the mark scheme and examiner's report
read the residual +80 at x = 32 and add it: w = −930 + 50.2(32) + 80 = 756
many forgot the residual
June 2022 · AS Paper 2 (Further Statistics 2) · Q3(e)1 mark
Mark scheme (whole question; this part is marked with the orange bar)
Examiner's comment
Answers to part (e) varied. Many correctly identified that the residuals were not ‘randomly scattered’ above and below 0 after x = 33 and that this meant the model may not be linear in this range. A common incorrect response was to assume that there was now ‘negative correlation’ since the residuals were decreasing as x increased.
Key points from the mark scheme and examiner's report
all 5 residuals for x > 33 are negative so the model systematically overestimates the weight of the longest bream (not random scatter)
"negative correlation of residuals" scores 0
June 2022 · Paper 4B (Further Statistics 2) · Q1(b)1 mark
Mark scheme (whole question; this part is marked with the orange bar)
Examiner's comment
This question proved to be an accessible start for virtually all candidates. Nearly all made a equation. Part (b) was generally well attempted with the majority of candidates giving a correct description of a residual. Incomplete answers simply stated how far away from the line of best fit without referencing the vertical/y distance.
Key points from the mark scheme and examiner's report
residual = observed value − predicted value of the dependent variable
'distance from the line' without saying vertical/y-distance scores 0
June 2022 · Paper 4B (Further Statistics 2) · Q1(c)2 marks
Mark scheme (whole question; this part is marked with the orange bar)
Examiner's comment
This question proved to be an accessible start for virtually all candidates. Nearly all made a Part (c) was again well attempted, though many lengthy attempts were seen at calculating the product moment correlation coefficient when candidates should have realised that they had all of the information to complete it in one step. It was not uncommon to see 2 significant figure answers given rather than 3 as instructed on the front cover of the examination paper.
Key points from the mark scheme and examiner's report
RSS = Syy(1 − r²) gives r = 0.246 in one step (or Sty = 45.5 × 52)
many took lengthy routes
give 3 s.f.
−0.246 scores M1A0
June 2022 · Paper 4B (Further Statistics 2) · Q1(d)2 marks
Mark scheme (whole question; this part is marked with the orange bar)
Examiner's comment
This question proved to be an accessible start for virtually all candidates. Nearly all made a (d)(i) was well accessed with most stating that there was weak correlation. Stating that the value of r was far from one was insufficient as candidates needed to make it clear that it is the proximity to 0 that makes it weak. Candidates struggled more in part (ii) to clearly express themselves, though a number of candidates did correctly identify that for values of t > 20 the residuals were no longer randomly scattered about 0.
Key points from the mark scheme and examiner's report
(i) r is close to 0, weak correlation ('far from 1' not enough)
(ii) for t > 20 the residuals are not randomly scattered about 0 / show a pattern
'not randomly scattered' alone scores 0
June 2022 · Paper 4B (Further Statistics 2) · Q1(e)1 mark
Mark scheme (whole question; this part is marked with the orange bar)
Examiner's comment
This question proved to be an accessible start for virtually all candidates. Nearly all made a Part (e) was the least successfully attempted part of the question with the majority of candidates suggesting that Kwame was indeed correct. Most did not realise that the value of RSS is not ‘standardised’ and depends on the size of the variable being measured.
Key points from the mark scheme and examiner's report
disagree: RSS is not standardised
the two RSS values are in different units/scales, so cannot compare strength of relationship
most wrongly agreed with Kwame
Nov 2021 · Paper 4B (Further Statistics 2) · Q4(d)1 mark
Mark scheme (whole question; this part is marked with the orange bar)
Key points from the mark scheme and examiner's report
RSS = Stt − Sxt²/Sxx = 444.7(1 − 0.959²) = 35.7
Nov 2021 · Paper 4B (Further Statistics 2) · Q4(e)(i)2 marks
Mark scheme (whole question; this part is marked with the orange bar)
Key points from the mark scheme and examiner's report
residual = 1.4 − (14.3 − 0.00681 × 1100)
squared = 29.2
29.2/35.7 = 82%
Nov 2021 · Paper 4B (Further Statistics 2) · Q4(e)(ii)1 mark
Mark scheme (whole question; this part is marked with the orange bar)
Key points from the mark scheme and examiner's report
this one point gives most of the RSS so it is possibly an outlier and should be investigated
Oct 2020 · AS Paper 2 (Further Statistics 2) · Q4(c)1 mark
Mark scheme (whole question; this part is marked with the orange bar)
Key points from the mark scheme and examiner's report
not appropriate: points follow a curve, so the line overestimates in the middle and underestimates at the ends
Oct 2020 · AS Paper 2 (Further Statistics 2) · Q4(d)2 marks
Mark scheme (whole question; this part is marked with the orange bar)
Key points from the mark scheme and examiner's report
Sww = Σw² − (Σw)²/18 = 22166.4 from the table
RSS = Sww(1 − r²) = 570
Oct 2020 · AS Paper 2 (Further Statistics 2) · Q4(e)1 mark
Mark scheme (whole question; this part is marked with the orange bar)
Key points from the mark scheme and examiner's report
strength depends on cross-sectional area, which is proportional to d²
Oct 2020 · AS Paper 2 (Further Statistics 2) · Q4(g)2 marks
Mark scheme (whole question; this part is marked with the orange bar)
Key points from the mark scheme and examiner's report
RSS = Sww − Swu²/Suu = 85.9 (r = 0.998)
Oct 2020 · AS Paper 2 (Further Statistics 2) · Q4(h)1 mark
Mark scheme (whole question; this part is marked with the orange bar)
Key points from the mark scheme and examiner's report
Robert's model is better because its RSS is much smaller (85.9 against 570)
Oct 2020 · Paper 4B (Further Statistics 2) · Q36 marks
Mark scheme (whole question; this part is marked with the orange bar)
Key points from the mark scheme and examiner's report
I: feasible but a systematic (wave) pattern suggests non-linear
II: impossible since residuals must sum to zero (all positive here)
III: feasible and randomly scattered about zero so linear
June 2019 · AS Paper 2 (Further Statistics 2) · Q3(e)2 marks
Mark scheme (whole question; this part is marked with the orange bar)
Examiner's comment
Part (e) was well answered with the majority of candidates gaining both marks.
Key points from the mark scheme and examiner's report
RSS = Sll − Swl²/Sww (or Sll(1 − r²)) = 0.256
June 2019 · AS Paper 2 (Further Statistics 2) · Q3(f)2 marks
Mark scheme (whole question; this part is marked with the orange bar)
Examiner's comment
Part (f) discriminated the most able candidates with a correct answer to both parts rarely being seen. Those candidates who made an attempt at answering were generally able to gain a mark in part(ii). The most common error in part (i) was to state that “all the points are close to zero” which may be true but does not necessarily mean a linear regression model is suitable. They need to be randomly scattered. The candidates who gained no marks in this part had left it unanswered. candidates should remember that leaving a part unanswered will definitely gain zero marks but if an attempt is made there is always a chance of scoring some marks.
Key points from the mark scheme and examiner's report
(i) residuals randomly scattered above and below 0, so linear model suitable ("all close to zero" scores 0)
(ii) remove the possible outlier (residual about 0.4) and recalculate the line
June 2019 · Paper 4B (Further Statistics 2) · Q2(e)2 marks
Mark scheme (whole question; this part is marked with the orange bar)
Examiner's comment
This question was well answered with the majority of candidates gaining the marks in the first four parts. Part (e) and (f) were less well answered. In part (e) the main misconception was that the residuals needed to be close to zero for the model to be suitable rather than they need to be randomly scattered. A minority of candidates stated that the residuals did not add up to 1 possibly having not noticed the residual for f = 0
Key points from the mark scheme and examiner's report
residuals show an 'n' shaped pattern / are not randomly scattered, so the linear model is unlikely to be suitable
main misconception was that residuals must be close to zero
June 2019 · Paper 4B (Further Statistics 2) · Q2(f)1 mark
Mark scheme (whole question; this part is marked with the orange bar)
Examiner's comment
This question was well answered with the majority of candidates gaining the marks in the first four parts. Part (e) and (f) were less well answered.
Key points from the mark scheme and examiner's report
fit a curve (e.g. quadratic) rather than a straight line
June 2018 · AS Paper 2 (Further Statistics 2) · Q1(c)2 marks
Mark scheme (whole question; this part is marked with the orange bar)
Examiner's comment
full marks in part (c).
Key points from the mark scheme and examiner's report
RSS = Spp − Smp²/Smm (or Spp(1 − r²)) = 42.3
June 2018 · AS Paper 2 (Further Statistics 2) · Q1(d)2 marks
Mark scheme (whole question; this part is marked with the orange bar)
Examiner's comment
Part (d) proved to be one of the most challenging parts of the entire paper as students did not realise that they could use the information on the residual plot to determine the exact physics score of the individual who scored 30 marks on the maths test. Most students simply opted to estimate it by using their linear regression equation, neglecting to add on the residual. Generally students were able to give a suitable reason in part (e) as to why the point with the
Key points from the mark scheme and examiner's report
read the residual (−1) off Figure 1 and ADD it to the predicted value: p = 1.08 + 0.604(30) + residual = 18
most just used the regression line and ignored the residual
one of the hardest parts
June 2018 · AS Paper 2 (Further Statistics 2) · Q1(e)1 mark
Mark scheme (whole question; this part is marked with the orange bar)
Examiner's comment
Generally students were able to give a suitable reason in part (e) as to why the point with the largest residual is removed before carrying out further analysis. The most common answer was to identify it as being an outlier. Part (f) was less successfully answered as it required
Key points from the mark scheme and examiner's report
its residual (about −5) is far from 0, so it may be an outlier / does not fit the trend
June 2018 · AS Paper 2 (Further Statistics 2) · Q1(f)1 mark
Mark scheme (whole question; this part is marked with the orange bar)
Examiner's comment
was to identify it as being an outlier. Part (f) was less successfully answered as it required both a statement about the relative size of the correlation coefficient and a correct supporting reason. Though some were able to identify that the new r would be closer to 1, many did not sufficiently justify why this was the case.
Key points from the mark scheme and examiner's report
r closer to 1 (increases) because the remaining points lie closer to a line
many gave the direction without a valid reason
Find an Edexcel Further Maths tutor
Stuck on this? Book a trial lesson with our Further Maths tutor.