Why linearise? When you suspect data follows y=axn or y=abx, taking logs converts the non-linear law to a STRAIGHT LINE. A scatter plot of the log-transformed data is easy to interpret: gradient and intercept give the unknown parameters.
Case 1: y=axn (power law).
Take ln of both sides:
lny=lna+nlnx.
This is Y=mX+c with Y=lny, X=lnx, m=n, c=lna.
Plot lny against lnx — log-LOG plot.
- Gradient = n.
- y-intercept (where lnx=0, i.e. x=1) = lna, so a=eintercept.
Worked example. Data (x,y) plotted with lny on the y-axis and lnx on the x-axis gives a line with gradient 1.5 and intercept ln2. Find the relationship.
- n=1.5.
- lna=ln2⇒a=2.
- Model: y=2x1.5.
Case 2: y=abx (exponential law).
Take ln of both sides:
lny=lna+xlnb.
This is Y=mX+c with Y=lny, X=x (NOT lnx!), m=lnb, c=lna.
Plot lny against x — log-LINEAR plot.
- Gradient = lnb, so b=egradient.
- y-intercept (where x=0) = lna, so a=eintercept.
Worked example. Data plotted gives a line with gradient 0.4 and y-intercept ln3. Find the model.
- lnb=0.4⇒b=e0.4≈1.492.
- lna=ln3⇒a=3.
- Model: y=3⋅1.492x.
Which axes to log?
| Model | Plot |
|---|
| y=axn | lny vs lnx |
| y=abx | lny vs x |
| y=a+bxn | not log-linear; can't be linearised directly |