Understanding the Total Sum of Squares in Finance

The total sum of squares is a foundational quantity in statistics and regression analysis — it measures the total variation in your data, and it's the starting point for working out how much of that variation a model can explain. Understanding it is essential to understanding R-squared, ANOVA and goodness of fit. This guide explains what the total sum of squares is, the formula, how it splits into explained and unexplained parts, and why it matters — in clear, plain language. It complements our guides to the F-distribution and the best linear unbiased estimator, and is relevant to anyone studying statistics or econometrics.

What is the total sum of squares?

The total sum of squares (often written TSS or SST) measures the total variation in a variable around its mean. It's calculated by taking each observation's deviation from the average, squaring it, and adding them all up:

TSS = Σ(y_i − ̅y)²

Here y_i is each data point and ̅y (y-bar) is the mean. Squaring the deviations ensures they're all positive and gives extra weight to larger deviations. The result is a single number capturing how spread out the data is overall — essentially the "total amount of variability" we're trying to understand or explain. (It's closely related to variance, which is just TSS divided by the number of observations, or degrees of freedom.)

Splitting variation: the key decomposition

The reason the total sum of squares is so important is that, in regression, it can be split into two parts:

TSS = ESS + RSS

where the explained sum of squares (ESS) is the variation captured by the regression model — the part the model accounts for — and the residual sum of squares (RSS) is the variation left over in the residuals, which the model fails to explain. In words: total variation = explained variation + unexplained variation. This decomposition is the heart of how we judge a model: it separates the variability the model handles from the variability it doesn't, and it holds exactly for ordinary least squares regression.

How it gives R-squared

This split leads directly to the most familiar measure of fit, R-squared. R-squared is simply the proportion of the total variation that the model explains:

R² = ESS ÷ TSS = 1 − (RSS ÷ TSS)

So if a model's explained sum of squares is 80% of the total sum of squares, R-squared is 0.8 — the model accounts for 80% of the variation, leaving 20% unexplained. This makes the total sum of squares the denominator against which a model's explanatory power is measured. Without TSS as the benchmark of total variability, R-squared would have nothing to be a fraction of.

The total sum of squares in ANOVA

The same idea appears in analysis of variance (ANOVA), where the total sum of squares is decomposed differently: into the variation between groups and the variation within groups. Comparing these (via an F-statistic and the F-distribution) tells us whether group means differ significantly. In both regression and ANOVA, the principle is identical: start with the total sum of squares, then partition it into meaningful components — explained versus unexplained, or between versus within — to draw conclusions.

Why the total sum of squares matters

The total sum of squares matters because it's the baseline against which explanation is measured. It quantifies how much variability there is to account for in the first place, and once you know that, every measure of model quality — R-squared, the F-test, ANOVA results — flows from comparing parts of it. It turns the vague idea of "how well does the model fit?" into a precise, decomposable calculation. That's why it's one of the first quantities computed in almost any regression or analysis-of-variance procedure.

Frequently asked questions

What is the total sum of squares?

A measure of the total variation in a variable around its mean: the sum of squared deviations, TSS = Σ(y_i − ̅y)².

How does it split up in regression?

TSS = ESS + RSS: total variation equals the explained sum of squares (captured by the model) plus the residual sum of squares (left unexplained in the residuals).

How is it related to R-squared?

R² = ESS ÷ TSS = 1 − RSS ÷ TSS — the proportion of total variation that the model explains. TSS is the benchmark of total variability.

How is it used in ANOVA?

The total sum of squares is split into between-group and within-group variation; comparing them with an F-test shows whether group means differ significantly.

Master statistics with Learnsignal

Quantities like the total sum of squares underpin data analysis. Learnsignal's tutor-led ACCA and CIMA courses build the quantitative foundations — with flexible, supported online study that fits around work.