A logarithmic regression model is useful when your data rises (or falls) quickly at first and then gradually levels out. This calculator fits the model:
y = a + b ln(x)
It estimates a (intercept) and b (slope on ln(x)) using least squares, then reports goodness-of-fit metrics so you can quickly judge whether a log model is a reasonable choice.
What is log regression?
Log regression (more precisely, logarithmic regression) describes a relationship where the response variable changes linearly with the natural log of the predictor. Instead of assuming straight-line growth with x, it assumes straight-line growth with ln(x).
- If b > 0, y increases as x increases, but with diminishing increments.
- If b < 0, y decreases as x increases.
- If b ≈ 0, x may not meaningfully explain y under this model.
When should you use a logarithmic model?
Choose this model when changes are fast early and slower later, such as learning curves, market saturation effects, and some biological adaptation processes.
Good practical signals
- Scatter plot looks curved but flattens over time.
- Plotting y versus ln(x) looks close to linear.
- You need a simple interpretable model rather than a high-complexity algorithm.
How to use this calculator
- Enter positive x values in order you measured them.
- Enter y values with the same number of points.
- Optionally enter a future x value for prediction.
- Click Calculate Regression.
The tool returns the equation, R², RMSE, sample size, and optional prediction value.
How the math works
We transform each x into z = ln(x), then run simple linear regression between z and y:
y = a + b z where z = ln(x)
From that, the final model is:
y = a + b ln(x)
This transformation keeps the model easy to estimate while capturing nonlinear behavior in the original x scale.
Interpreting results correctly
Coefficient b
b is the expected change in y for a one-unit increase in ln(x), not a one-unit increase in x directly. In plain language, equal percentage-like expansions in x can have similar marginal impacts on y.
R² and RMSE
- R²: Proportion of variance explained by the model (closer to 1 is better).
- RMSE: Typical prediction error size in y-units (smaller is better).
Common mistakes to avoid
- Using zero or negative x values (ln is undefined there).
- Comparing models solely by R² without inspecting residual behavior.
- Extrapolating far beyond observed x values.
- Confusing logarithmic regression with logistic regression.
Quick FAQ
Is this the same as logistic regression?
No. Logistic regression is for classification probabilities. This calculator performs logarithmic curve fitting for continuous y outcomes.
Which log base is used?
Natural log (base e) is used. If needed, other log bases can be converted by a constant scaling factor in the slope.
Can I use this for forecasting?
Yes, for cautious short-range forecasting when your process reasonably follows diminishing returns. Always validate with out-of-sample checks when possible.