Understanding Dot Notation in Plain English
Learn how to interpret row totals, column totals, grand totals, squared totals, and sums of squared observations without getting lost in statistical notation.
What each subscript and dot means.
“Sum then square” from “square then sum.”
Dot notation in ANOVA and DOE formulas.
1. The Basic Idea
Suppose an observation is written as:
Here, i usually identifies the row, treatment, or factor level, while j identifies the observation, replicate, or column.
2. How to Read Each Form
| Notation | Meaning | How to read it |
|---|---|---|
| \(y_{ij}\) | One individual observation | Observation j in row or treatment i |
| \(y_{i\cdot}\) | Sum across all j values for row i | Row i total |
| \(y_{\cdot j}\) | Sum across all i values for column j | Column j total |
| \(y_{\cdot\cdot}\) | Sum of every observation | Grand total |
Row 1, all columns
All rows, column 2
All rows, all columns
Sum over that position
Three indices: factor A, factor B, and replicate
In a replicated two-factor experiment, write a measurement as \(y_{ijk}\). Define the positions first: \(i=1,\ldots,a\) identifies factor A, \(j=1,\ldots,b\) identifies factor B, and \(k=1,\ldots,n\) identifies a replicate within that A–B combination. Here, every combination has the same number \(n\) of observations. A cell means one combination of factor levels.
The plain-text form y.j. means \(y_{\cdot j\cdot}\): keep B level j fixed; add over every A level and every replicate. It has three subscript positions. It is different from the two-position \(y_{\cdot j}\). The dots are placeholders for summed indices, not decimal points or multiplication signs.
| Notation / typed form | Held fixed | Explicit summation | Number of values added |
|---|---|---|---|
\(y_{ijk}\) / yijk | i, j, k | One observation; no summation | 1 |
\(y_{ij\cdot}\) / yij. | i, j | \(\sum_{k=1}^{n}y_{ijk}\) | n |
\(y_{i\cdot k}\) / yi.k | i, k | \(\sum_{j=1}^{b}y_{ijk}\) | b |
\(y_{\cdot jk}\) / y.jk | j, k | \(\sum_{i=1}^{a}y_{ijk}\) | a |
\(y_{i\cdot\cdot}\) / yi.. | i | \(\sum_{j=1}^{b}\sum_{k=1}^{n}y_{ijk}\) | bn |
\(y_{\cdot j\cdot}\) / y.j. | j | \(\sum_{i=1}^{a}\sum_{k=1}^{n}y_{ijk}\) | an |
\(y_{\cdot\cdot k}\) / y..k | k | \(\sum_{i=1}^{a}\sum_{j=1}^{b}y_{ijk}\) | ab |
\(y_{\cdot\cdot\cdot}\) / y... | Nothing | \(\sum_{i=1}^{a}\sum_{j=1}^{b}\sum_{k=1}^{n}y_{ijk}\) | abn |
The forms that retain k are valid sums, but a replicate number alone does not automatically represent a common block, time, or matched subject across cells. Give k that interpretation only when the experimental design supports it.
A bar changes a total into a mean
In this lesson, an unbarred y with dots is a total; a bar indicates a mean. Dots tell you which positions were aggregated. The bar tells you that division by the number of included observations is also required.
3. Numerical Example
| Treatment | Observation 1 | Observation 2 | Observation 3 | Row total |
|---|---|---|---|---|
| 1 | 10 | 12 | 11 | 33 |
| 2 | 15 | 16 | 17 | 48 |
| 3 | 20 | 21 | 22 | 63 |
4. What Happens When There Is a Square?
Sum first, then square
This is a squared total.
Square first, then sum
This is the sum of squared observations.
These two expressions are not interchangeable.
5. Why They Are Different
When a total is squared, cross-product terms are created:
That is why the squared total, 1089, is much larger than the sum of squares, 365.
Dot notation versus the summation sign
Dot notation names an aggregated quantity. The summation sign \(\sum\) writes out the addition operation. They can represent exactly the same total; they are not competing statistical methods.
On the left, two dots say which indices have been summed out. On the right, the two summation signs explicitly give their indices and limits. The j remains fixed on both sides.
| Use dot notation when… | Use a summation sign when… |
|---|---|
| Referring repeatedly to a row, column, cell, or grand total. | Showing precisely which values to add and their bounds. |
| Keeping a statistical formula compact. | Adding transformed terms, such as squares, residuals, or chi-square contributions. |
| Showing the indices retained after aggregation. | Deriving or implementing the calculation step by step. |
Why one formula can contain both
- For a fixed j, add all values across i and k.
- Square that B-level total.
- Repeat for each j, then add those squared totals.
The outer sum still has a job: the dot total retains j. By contrast, \(\sum_{i,j,k}y_{ijk}^{2}\) squares each individual observation before adding, and \(y_{\cdot\cdot\cdot}^{2}\) squares the single grand total. These three operations generally give different results.
6. Interactive Practice
Enter three observations from one treatment row.
7. Where This Appears in ANOVA and DOE
Dot notation is used to shorten formulas for treatment totals, block totals, correction factors, and sums of squares.
| Quantity | Typical notation | Interpretation |
|---|---|---|
| Treatment total | \(y_{i\cdot}\) | All observations in treatment i added together |
| Block total | \(y_{\cdot j}\) | All observations in block j added together |
| Grand total | \(y_{\cdot\cdot}\) | All observations added together |
| Correction factor | \(y_{\cdot\cdot}^{2}\)/N | Grand total squared, then divided by total sample size |
| Raw sum of squares | \(\displaystyle \sum_i\sum_j y_{ij}^{2}\) | Square every observation, then add them |
Worked example: dot notation in two-factor ANOVA
Use this small, synthetic strength dataset (arbitrary units). A represents two temperature settings; B represents two cooling settings. Each cell contains two independent experimental runs, so \(a=b=n=2\) and \(N=abn=8\). The purpose is to practise the notation, not recommend process settings.
| A level | B1: k1, k2 | B2: k1, k2 | A-level total |
|---|---|---|---|
| A1 | 8, 10 (cell total 18) | 12, 14 (cell total 26) | 44 |
| A2 | 10, 12 (cell total 22) | 18, 20 (cell total 38) | 60 |
| B-level totals | 40 | 64 | Grand total 104 |
Thus \(\bar y_{\cdot2\cdot}=64/(an)=64/4=16\), while \(\bar y_{\cdot\cdot\cdot}=104/8=13\). The first is the mean at B2; the second is the mean of all eight measurements.
Reading the sums-of-squares formulas
For this balanced, replicated, fixed-effects two-factor model, define the correction factor \(C=y_{\cdot\cdot\cdot}^{2}/(abn)=104^2/8=1352\). The raw sum of squared observations is \(8^2+10^2+12^2+14^2+10^2+12^2+18^2+20^2=1472\).
Here T means total corrected variation, A and B are the main effects, AB is their interaction, and E is within-cell error. Check the result directly: every observation is one unit from its cell mean, so eight squared deviations of 1 give \(SS_E=8\). The correction factor removes the overall mean contribution; it is not an error correction or a chi-square continuity correction.
| Source | SS | Degrees of freedom | MS = SS / df | F = effect MS / error MS |
|---|---|---|---|---|
| A | 32 | a − 1 = 1 | 32 | 16 |
| B | 72 | b − 1 = 1 | 72 | 36 |
| A × B | 8 | (a − 1)(b − 1) = 1 | 8 | 4 |
| Error | 8 | ab(n − 1) = 4 | 2 | — |
| Total | 120 | abn − 1 = 7 | — | — |
ANOVA compares variation between factor-level means with residual variation. The observed B2–B1 difference is 4 at A1 and 8 at A2: this difference of differences explains the nonzero interaction SS. Statistical significance requires the appropriate F distribution and degrees of freedom, not just comparing raw SS values.
Model and decomposition reference: NIST: two-way ANOVA. The dataset and arithmetic here are original teaching examples.
Worked example: chi-square test of independence
Now the cells contain counts of items, not strength measurements. In this synthetic example, 100 independently sampled pipes are classified by production line and inspection result. Each pipe is counted once. Let \(O_{ij}\) be the observed count in row i and column j. Some books write \(n_{ij}\) instead.
| Line (i) | Pass (j = 1) | Fail (j = 2) | Row total \(O_{i\cdot}\) |
|---|---|---|---|
| Line 1 | 30 | 20 | 50 |
| Line 2 | 45 | 5 | 50 |
| Column total \(O_{\cdot j}\) | 75 | 25 | \(O_{\cdot\cdot}=100\) |
The null hypothesis is that line and inspection result are independent. Under independence, each line would have the overall pass proportion, 75/100, and fail proportion, 25/100. Multiply the row size by these proportions to obtain expected counts:
Line 2 has the same row total, so its expected counts are also 37.5 and 12.5. Expected counts can be fractional. Their row totals, column totals, and grand total match those of the observed table.
The uncorrected Pearson test gives p ≈ 0.000532. At α = 0.05, reject independence: there is evidence of an association between line and inspection result. This does not establish that the line caused the difference.
Use nonnegative frequency counts, mutually exclusive categories, independent observations, and adequately large expected counts. All four expected counts here exceed 5. For sparse tables, consider a suitable exact method rather than relying on the chi-square approximation. This example uses no Yates continuity correction.
| Question | ANOVA | Chi-square independence |
|---|---|---|
| What is recorded? | A numerical response for each experimental unit | A count for each pair of categories |
| What do dots summarize? | Response totals or, with a bar, means | Row counts, column counts, and sample size |
| What does Σ combine? | Squared totals or squared deviations, as the formula specifies | Standardized squared observed–expected differences |
| What is tested? | Mean effects and interaction in the specified model | Association between two categorical variables |
Method and implementation references: Microsoft: CHISQ.TEST and Minitab: table analyses.
8. Excel Implementation
| Task | Excel formula | Use case |
|---|---|---|
| Row total | =SUM(B2:D2) | Equivalent to \(y_{i\cdot}\) |
| Squared row total | =SUM(B2:D2)^2 | Equivalent to \(y_{i\cdot}^{2}\) |
| Sum of squared observations | =SUMSQ(B2:D2) | Equivalent to \(\displaystyle \sum_j y_{ij}^{2}\) |
| Grand total | =SUM(B2:D4) | Equivalent to \(y_{\cdot\cdot}\) |
| Grand total squared | =SUM(B2:D4)^2 | Equivalent to \(y_{\cdot\cdot}^{2}\) |
SUM(range)^2 and SUMSQ(range) calculate different quantities.9. Minitab Connection
Minitab normally performs these calculations internally when you run an ANOVA.
| Analysis | Minitab path |
|---|---|
| One-way ANOVA | Stat → ANOVA → One-Way |
| General factorial ANOVA | Stat → ANOVA → General Linear Model → Fit General Linear Model |
| Factorial DOE analysis | Stat → DOE → Factorial → Analyze Factorial Design |
For exams, remember that the notation explains the manual calculations, while Minitab reports the resulting sums of squares, mean squares, F-statistics, and p-values.
Statistics Implementation
Excel Functions
| Function | Syntax | Purpose | When to use it |
|---|---|---|---|
| SUM | SUM(number1, [number2], …) | Add values | Cell and marginal totals |
| SUMSQ | SUMSQ(number1, [number2], …) | Add squared values | Raw sum of squares, or squared totals when the input cells already contain totals |
| AVERAGE | AVERAGE(number1, [number2], …) | Arithmetic mean | Barred y notation for numeric observations |
| SUMPRODUCT | SUMPRODUCT(array1, [array2], …) | Sum array calculations | Combine chi-square cell contributions |
| CHISQ.TEST | CHISQ.TEST(actual_range, expected_range) | Return the chi-square p-value | Observed and expected count tables of matching shape |
| CHISQ.DIST.RT | CHISQ.DIST.RT(x, deg_freedom) | Right-tail probability | Get p from an already calculated chi-square statistic |
| F.DIST.RT | F.DIST.RT(x, deg_freedom1, deg_freedom2) | Right-tail F probability | Get p from an ANOVA F ratio and its numerator/error df |
The right-tail variants are used for these tests. CHISQ.DIST(x,df,TRUE) and F.DIST(x,df1,df2,TRUE) return left-tail cumulative probabilities instead; the FALSE versions return densities, not p-values.
Excel Use Cases
ANOVA data: enter the four A–B cells as rows 2–5, with the two replicate values in B:C: (8,10), (12,14), (10,12), (18,20). Put =SUM(B2:C2) in D2 and fill to D5. Then:
- Cell total: D2 = 18; cell mean:
=AVERAGE(B2:C2)= 9. - A1 total:
=SUM(D2:D3)= 44; B2 total:=SUM(D3,D5)= 64. - Grand total:
=SUM(B2:C5)= 104; raw sum of squares:=SUMSQ(B2:C5)= 1472. - Total SS:
=SUMSQ(B2:C5)-SUM(B2:C5)^2/8= 120. - Sum of squared cell totals:
=SUMSQ(D2:D5)= 2928. These inputs are already totals, unlike the raw observations in B:C. - Interaction p-value:
=F.DIST.RT(4,1,4)≈ 0.1161. The nonzero observed interaction is not statistically significant at 0.05 in this tiny example.
Chi-square data, on a separate sheet: place 30,20 in B2:C2 and 45,5 in B3:C3. Row totals go in D2:D3, column totals in B4:C4, and the grand total in D4. Put =$D2*B$4/$D$4 in F2, then fill across to G2 and down to row 3. Expected counts in F2:G3 are 37.5,12.5 in both rows.
- Statistic:
=SUMPRODUCT((B2:C3-F2:G3)^2/F2:G3)= 12. - P-value:
=CHISQ.TEST(B2:C3,F2:G3)≈ 0.000532. - Check:
=CHISQ.DIST.RT(12,1)gives the same p-value.
Exclude totals from the test ranges. CHISQ.TEST returns a p-value, not the statistic 12. These references assume the exact layouts above and no missing observations.
Minitab Navigation
For the ANOVA example, use eight worksheet rows with columns for response, A, and B. Choose Stat → ANOVA → General Linear Model → Fit General Linear Model; enter the response and the two categorical factors, then include A, B, and A*B in the model. Use Stat → ANOVA → One-Way only for a one-factor question. For a design already defined in Minitab's DOE tools, use Stat → DOE → Factorial → Analyze Factorial Design.
For counts, choose Stat → Tables → Chi-Square Test for Association and Summarized data in a two-way table; supply only the count columns, without totals. Alternatively, Stat → Tables → Cross Tabulation and Chi-Square also provides cross-tabulations and the independence test. Use the Pearson result to match this example.
Exam Tips
- For ASQ CSSBB/CQE-style calculations, identify what each index means before substituting numbers.
- Count the observations inside each total before selecting the denominator.
- Distinguish a squared total, a sum of squares, a mean square, and a test statistic.
- For independence, use (rows − 1)(columns − 1), exclude margins, and work with counts.
- In practice, confirm the design and assumptions before interpreting software p-values.
10. Quick Knowledge Check
Which expression represents \(10^2+12^2+11^2\)?
11. Final Summary
| Expression | Operation order | Plain-English meaning |
|---|---|---|
| \(y_{i\cdot}\) | Sum | Add all observations in row i |
| \(y_{i\cdot}^{2}\) | Sum, then square | Square the row total |
| \(\displaystyle \sum_j y_{ij}^{2}\) | Square, then sum | Square each observation in row i and add them |
| \(y_{\cdot\cdot}\) | Sum everything | Grand total |
| \(\displaystyle \sum_i\sum_j y_{ij}^{2}\) | Square every observation, then sum | Raw sum of squares for the complete dataset |
Knowledge check
Check your understanding
Comprehension quiz
Answer all eight questions, then submit for your score and explanations. You can change answers and submit again.