The University of Manchester · FACULTY OF ENVIRONMENTAL SCIENCE

EART60061 Chap.7 Regression, Significance and Power-Law Rating Curves

- one subject, every graph, every model, every mark
7 Chapters4-page Bible
Our own words - no uploaded lecturer files
Updated for this semester
Chapter 7 of 10 · EART60061

Regression, Significance and Power-Law Rating Curves

Regression and Rating Curves treats a relationship between two variables as a model to be tested. Regression gives the equation of the most likely straight line and r squared the share of variation explained, but significance depends on the number of points, so r is converted to t with n minus 2 degrees of freedom and then to a p value.

Dilution implies that concentration is proportional to discharge raised to the power minus one, a power law that straightens on a double-log plot, whereas exponential relationships straighten on a semi-log plot.

The catchment case study improves a weak concentration and discharge relationship step by step: log transformation, splitting rising and falling limbs to capture solute depletion, and removing the hydrogen ion contribution of acidic rain.

The case study also shows that a model can succeed in an engineering sense, predicting concentration from discharge, while its greater value lies in what it reveals about the sources of solute in the catchment. Throughout, every r squared is reported with its p value, because the number of points controls whether a fit means anything.

In this chapter

What this chapter covers

  • 01

    Regression and correlation

  • 02

    From r squared to p

  • 03

    Power laws and double-log plots

  • 04

    Exponentials and semi-log plots

  • 05

    Rising and falling limbs

  • 06

    Separating hydrogen ions from total solutes

  • 07

    Engineering success versus process understanding

Worked example · free

Decide whether a sulphate and discharge regression is significant

Q [5 marks]. Fifteen paired samples give r squared = 0.30 for sulphate against discharge. Calculate t and p and state whether the regression is significant at 95% confidence. The marks shown are AskSia's, not a published University scheme.
Concentration is plotted against discharge with points joined in time order; the falling limb lies below the rising limb at every discharge, so the points traceConcentration is plotted against discharge with points joined in time order; the falling limb lies below the rising limb at every discharge, so the points trace a loop rather than scattering randomly about one line.020406004080Discharge QConcentration Crising limbfalling limbpeak discharge
Concentration is plotted against discharge with points joined in time order; the falling limb lies below the rising limb at every discharge, so the points trace a loop rather than scattering randomly about one line.
  • 1r = square root of 0.30 = 0.548; degrees of freedom = n minus 2 = 13.
  • 2t = 0.548 times square root of 13, divided by square root of 0.70: 0.548 times 3.606 / 0.837 = 2.36.
  • 2TDIST(2.36, 13, 2) gives p = 0.035, which is below 0.05, so the regression is significant.
t = 2.36 and p = 0.035, so the relationship is significant at 95% confidence even though discharge explains only 30% of the variation in sulphate.
Sia tip — Report r squared and p together; neither a high r squared nor a significant p tells the whole story alone.
Glossary

Key terms

Power law
A relationship of the form y equals x raised to a constant power, which becomes a straight line on a double-log plot.
Rating curve
A relationship used to predict a river quantity, here solute concentration, from discharge.
Electrical conductivity
A measure of all dissolved ions in water, used as an aggregate indicator of solute concentration.
Hydrograph limb
The rising or falling part of a discharge record during a storm event, used to split data in time.
Double-log plot
A graph of the logarithm of y against the logarithm of x, on which a power law appears as a straight line.
FAQ

Regression, Significance and Power-Law Rating Curves FAQ

Why is a straight line the wrong model for concentration against discharge?

If dilution controls concentration, adding water lowers concentration but can never reduce it to zero, while a straight line with a negative slope eventually crosses the axis. Dilution makes concentration proportional to one over volume, a power law, so the data should be tested on a double-log plot.

Why can a high r squared fail to be significant?

With only two points a line always fits perfectly, and with very few points a high r squared can arise by chance. Significance depends on the number of points through the degrees of freedom, so four points with r squared of 0.85 can still give p above 0.05.

Why does splitting the data into rising and falling limbs help?

Solute is more available before a storm peak and depleted after it, so the relationship between concentration and discharge changes over time. Splitting the data builds that time dependence into the model and improves the fit, although fewer points per limb means significance must be rechecked.

How do I calculate p from r squared?

Take the square root of r squared to get r, then compute t as r times the square root of n minus 2, divided by the square root of one minus r squared. Use the t distribution with n minus 2 degrees of freedom and two tails; a p value below 0.05 is significant at 95%.

Why subtract hydrogen ions from conductivity?

Conductivity measures all dissolved ions, but hydrogen ions arrive largely in rain while other ions come from weathering rock. Converting pH to hydrogen ion concentration and removing its share separates the two sources, and on the falling limb this raised r squared from 0.448 to 0.979.

Study strategy

Exam move

Set up a small p-from-r calculator in Excel with cells for r squared and degrees of freedom, and use it on every regression you make. Practise deciding which plot straightens which law before plotting. Then retell the catchment case study as a sequence of model improvements, naming the assumption relaxed at each step and the effect on r squared and p.

Write each model improvement as one line: what you changed, why the process justified it, and the new r squared and p.

A+Everything unlocked
Unlocks this Bible + all 41 of your subjects at The University of Manchester, plus 1,000+ more Bibles across the AskSia library.
Sia - your EART60061 tutor, unlimited, worked the way the exam marks it
The full 4-page Bible + practice bank with worked solutions
Chrome extension - sync your LMS so Sia knows your deadlines
Bilingual EN / Chinese on every Bible and every Sia answer
$0.99 Trial
30-day money-back · cancel in one tap · how it works