Data Analysis Formula Sheet and Short Notes
One-page CSIR NET Chemical Sciences revision: mean, SD, RSD, confidence limits, errors and their propagation, least-squares lines, covariance and correlation.
- 6 min
On this page · 6 sections
Every formula and trap from Mean and Standard Deviation, Absolute and Relative Errors and Linear Regression, Covariance and Correlation, on one page.
Chemical Sciences marking
Per the NTA information bulletin for December 2026:
| Part | Given | Attempt | Correct | Wrong | Maximum |
|---|---|---|---|---|---|
| B | 40 | 35 | +2 | −0.5 | 70 |
| C | 60 | 25 | +4 | −1 | 100 |
Only the first 35 (Part B) and 25 (Part C) answered questions are evaluated. See the chapter overview.
Mean and standard deviation
$$\boxed{\bar{x} = \frac{\sum x_i}{n}} \qquad \boxed{s = \sqrt{\frac{\sum (x_i - \bar{x})^2}{n - 1}} = \sqrt{\frac{\sum x_i^2 - (\sum x_i)^2/n}{n - 1}}}$$- \( \sum (x_i - \bar{x}) = 0 \) always. Population \( \sigma \) divides by \( N \) about \( \mu \).
- RSD = \( s/\bar{x} \); CV = RSD × 100%; ppt = × 1000.
- For \( y = ax + b \): mean → \( a\bar{x} + b \); SD → \( \lvert a \rvert s \); variance → \( a^2 s^2 \); RSD changes if \( b \ne 0 \).
- Pooled SD: \( \sqrt{\sum (n_j - 1)s_j^2 / \sum (n_j - 1)} \).
| Gaussian band | ±1σ | ±1.96σ | ±2σ | ±3σ |
|---|---|---|---|---|
| Inside | 68.3% | 95.0% | 95.4% | 99.7% |
| d.f. | 1 | 2 | 3 | 4 | 5 | ∞ |
|---|---|---|---|---|---|---|
| \( t \) (95%) | 12.71 | 4.30 | 3.18 | 2.78 | 2.57 | 1.96 |
- Four times the replicates halves the standard error.
- t-test vs known value: \( t = \lvert \bar{x} - \mu \rvert \sqrt{n}/s \). F-test: \( F = s_1^2/s_2^2 \), larger on top.
Absolute and relative errors
$$\boxed{E_a = x - x_t} \qquad \boxed{E_r = \frac{x - x_t}{x_t}}$$| Accuracy | Precision | |
|---|---|---|
| Closeness to | true value | each other |
| Measured by | error | SD, RSD |
| Spoilt by | systematic error | random error |
- Systematic: instrumental, method, personal; one sign; not reduced by repeats; found with reference materials, blanks, independent methods, varied sample size.
- Constant error: relative size falls with larger samples. Proportional: relative size constant.
- Random: both signs, Gaussian, falls as \( 1/\sqrt{n} \).
- Sig figs: add/subtract by decimal places; multiply/divide by significant figures; decimals in a log = sig figs in the number.
Propagation
| \( y \) | Maximum error | Standard deviation |
|---|---|---|
| \( a \pm b \) | \( E_a + E_b \) | \( \sqrt{s_a^2 + s_b^2} \) |
| \( ab \), \( a/b \) | \( \frac{E_a}{a} + \frac{E_b}{b} \) (relative) | \( \sqrt{(s_a/a)^2 + (s_b/b)^2} \) (relative) |
| \( a^n \) | \( n\,E_a/a \) (relative) | \( n\,s_a/a \) (relative) |
| \( \log_{10} a \) | \( 0.434\,E_a/a \) | \( 0.434\,s_a/a \) |
| \( \ln a \) | \( E_a/a \) | \( s_a/a \) |
| \( 10^a \) | \( 2.303\,E_a \) (relative) | \( 2.303\,s_a \) (relative) |
- ±0.01 pH ≈ ±2.3% in [H⁺]; ±0.02 pH ≈ ±4.6%.
- \( E_a = 50 \) kJ mol⁻¹ at 300 K: 1 K error ≈ 6.7% in \( k \).
- The largest relative term dominates a quadrature sum.
Linear regression
$$\boxed{S_{xx} = \sum x^2 - \frac{(\sum x)^2}{n}} \qquad \boxed{S_{xy} = \sum xy - \frac{\sum x \sum y}{n}} \qquad \boxed{S_{yy} = \sum y^2 - \frac{(\sum y)^2}{n}}$$$$\boxed{b = \frac{S_{xy}}{S_{xx}}} \qquad \boxed{a = \bar{y} - b\bar{x}} \qquad \boxed{x_0 = \frac{y_0 - a}{b}}$$- The line passes through \( (\bar{x}, \bar{y}) \). Through the origin: \( b = \sum xy/\sum x^2 \).
- \( s_{y/x} = \sqrt{\sum e_i^2/(n - 2)} \); \( s_b = s_{y/x}/\sqrt{S_{xx}} \); \( s_a = s_{y/x}\sqrt{\sum x^2/(nS_{xx})} \).
- Wide spread of \( x \) → precise slope. Predict near \( \bar{x} \); do not extrapolate.
- Curved residuals → wrong model; fan-shaped → weighting needed.
| Law | Plot | Slope |
|---|---|---|
| First order | \( \ln[\text{A}] \) vs \( t \) | \( -k \) |
| Second order | \( 1/[\text{A}] \) vs \( t \) | \( k \) |
| Arrhenius | \( \ln k \) vs \( 1/T \) | \( -E_a/R \) |
| van ’t Hoff | \( \ln K \) vs \( 1/T \) | \( -\Delta H^\circ/R \) |
| Clausius–Clapeyron | \( \ln p \) vs \( 1/T \) | \( -\Delta_{\text{vap}}H/R \) |
| Lineweaver–Burk | \( 1/v \) vs \( 1/[\text{S}] \) | \( K_M/V_{\max} \) |
| Stern–Volmer | \( F_0/F \) vs \( [\text{Q}] \) | \( K_{SV} \) |
| Kohlrausch | \( \Lambda_m \) vs \( \sqrt{c} \) | \( -K \); intercept \( \Lambda_m^\circ \) |
With \( \log_{10} \) instead of \( \ln \), divide the slope by 2.303.
Covariance and correlation
$$\boxed{\text{cov}(x, y) = \frac{S_{xy}}{n - 1}} \qquad \boxed{r = \frac{S_{xy}}{\sqrt{S_{xx}S_{yy}}}} \qquad \boxed{b_{yx} = r\frac{s_y}{s_x},\ \ b_{xy} = r\frac{s_x}{s_y},\ \ b_{yx}b_{xy} = r^2}$$- Covariance has units and depends on scale: \( \text{cov}(px + q, uy + v) = pu\,\text{cov}(x, y) \).
- \( -1 \le r \le 1 \); \( r \), \( b_{yx} \), \( b_{xy} \) share a sign; \( r \) unchanged by change of units or origin, sign flips if one variable is multiplied by a negative number.
- \( r^2 \) = fraction of variance explained (\( r = 0.9 \) → 81%).
- Both regression lines pass through \( (\bar{x}, \bar{y}) \): solve them together for the means. Slope product > 1 means you have the lines the wrong way round.
- \( \text{var}(x \pm y) = s_x^2 + s_y^2 \pm 2\,\text{cov}(x, y) \).
- \( r = 0 \) for \( y = x^2 \) on symmetric \( x \); \( r = 0.981 \) for \( y = x^2 \), \( x = 1 \) to 5. \( r \) sees straight lines only.
Last-minute checks
Related: Spectroscopic Methods of Analysis for calibration and detection limits, and Radio-analytical Techniques for counting statistics.