18.2 Error Analysis: Systematic vs Random Errors & Percent Error

18.2 Error Analysis: Systematic vs Random Errors & Percent Error

Quick Summary: Experimental chemistry requires rigorous quantitative evaluation of data reliability. Accuracy reflects agreement with the true literature value, while precision reflects reproducibility among independent trials. Systematic (determinate) errors produce unidirectional bias (always too high or always too low) from uncalibrated glassware, instrumental zero offsets, or incomplete reactions; they compromise accuracy and can be detected and eliminated. Random (indeterminate) errors cause bidirectional statistical scatter governed by Gaussian distributions; they degrade precision and are minimized by averaging replicate measurements ($s_{\bar{x}} = s / \sqrt{N}$). Quantitative error is reported as percent error. In mathematical operations, absolute uncertainties combine in quadrature for addition/subtraction, while relative percentage uncertainties combine in quadrature for multiplication/division.

---

1. Accuracy versus Precision & The Target Paradigm

Every physical measurement possesses an inherent degree of uncertainty. In analytical and general chemistry, data quality is evaluated using two fundamentally distinct criteria:

• Accuracy: The closeness of agreement between an experimental value (or the mean of replicate values, $\bar{x}$) and the true, accepted reference value ($x_{\text{true}}$). Accuracy reflects the correctness of the analytical result and is degraded primarily by systematic errors.
• Precision: The closeness of agreement among independent experimental measurements obtained under stipulated identical conditions (repeatability and reproducibility). Precision reflects the scatter or dispersion of data points around their central mean and is limited primarily by random errors.

The Target Analogy

The classical target paradigm clarifies the four possible combinations of experimental data quality:

1. High Accuracy, High Precision: All darts cluster tightly in the center bullseye. The replicate measurements agree closely with one another, and their mean matches the accepted reference value.
2. Low Accuracy, High Precision: All darts cluster tightly together, but far away from the bullseye in an off-center ring. This pattern is the hallmark of a systematic error (e.g., an uncalibrated balance with a zero offset). Replicates are reproducible, but uniformly wrong.
3. High Accuracy, Low Precision: Darts are widely scattered across the target, but distributed symmetrically around the center bullseye. The average position matches the bullseye, but individual measurements exhibit substantial random scatter.
4. Low Accuracy, Low Precision: Darts are widely scattered across the target and displaced off to one side. The data suffers from both severe random noise and uncorrected systematic bias.

Comparison: Accuracy vs. Precision

---

2. Classification of Experimental Errors

Experimental discrepancies are partitioned into three distinct categories based on their origin and mathematical behavior:

Systematic (Determinate) Errors

Systematic errors possess an identifiable cause and impart a unidirectional bias to experimental results—measurements are consistently too high or consistently too low. They shift the sample mean away from the true value without necessarily altering the spread among replicates.

• Instrumental Errors: Caused by imperfections, wear, or improper calibration of measuring instruments.
  • Example 1: An electronic balance tared incorrectly with a $+0.0250\text{ g}$ positive zero offset adds $0.0250\text{ g}$ to every mass reading.
  • Example 2: Volumetric glassware (e.g., a $50.00\text{ mL}$ volumetric flask) calibrated at $20^\circ\text{C}$ used at $35^\circ\text{C}$; thermal expansion of both glass and solvent alters delivered volume.
  • Example 3: An uncalibrated pH meter whose glass electrode reads consistently $0.3\text{ pH}$ units low across all buffer standards.
• Methodic Errors: Arise from physical or chemical non-idealities inherent in the chosen analytical procedure.
  • Example 1: Gravimetric analysis in which the precipitate possesses a slight finite solubility ($K_{sp} > 0$) in the wash solvent, systematically decreasing collected mass.
  • Example 2: Co-precipitation or occluded adsorption of foreign counterions within crystal lattices, systematically inflating precipitate mass.
  • Example 3: In acid-base titrations, selecting an indicator whose visual color transition (endpoint) occurs at a pH significantly different from the theoretical stoichiometric equivalence point (titration error).
• Personal / Operative Errors: Result from unconscious bias or consistent procedural blunders by the human analyst.
  • Example 1 (Parallax Error): Consistently reading a buret meniscus from above eye level makes the meniscus appear aligned with a higher graduation, so each buret reading is recorded too low (buret numbers increase downward). In a graduated cylinder, the same habit gives readings that are too high.
  • Example 2: Consistently under-titrating by stopping at the very first transient flash of color rather than waiting for persistent color change.
• Detection and Elimination: Systematic errors can be discovered and corrected by:
  1. Calibrating instruments against NIST-traceable reference standards.
  2. Running reagent blanks (measuring background signals from solvent and reagents in the absence of analyte).
  3. Analyzing certified Standard Reference Materials (SRMs) of known composition.
  4. Employing independent secondary analytical techniques based on different physical principles (e.g., confirming atomic absorption spectrophotometry via gravimetric analysis).

Random (Indeterminate) Errors

Random errors arise from unpredictable, uncontrollable fluctuations in experimental conditions that occur at the microscopic level:

• Sources: Thermal convection currents inside an analytical balance draft shield, line-voltage micro-fluctuations in spectrophotometer power supplies, minor temperature changes during a kinetics run, and human visual estimation of the final uncertain decimal digit between analog scale graduations.
• Mathematical Characteristics: Random errors follow a continuous Gaussian (normal) distribution curve centered on the mean. Positive and negative deviations occur with equal statistical probability, and small deviations occur far more frequently than large deviations.
• Mitigation via the Central Limit Theorem: While random errors cannot be eliminated individually, their effect on experimental accuracy diminishes as replicate measurements increase. The standard error of the mean ($s_{\bar{x}}$) decreases inversely with the square root of the number of trials ($N$): $s_{\bar{x}} = \frac{s}{\sqrt{N}}$ Increasing the number of trials from $N = 1$ to $N = 4$ reduces the random uncertainty of the mean by half ($50\%$).

Gross Personal Errors (Blunders)

Gross errors are catastrophic singular failures that completely invalidate an individual experimental run (e.g., dropping a crucible, spilling a portion of analyte during transfer, or misrecording a balance display by an entire digit). Data points contaminated by gross errors are statistical outliers. They may be identified and rejected using formalized statistical rejection criteria (such as the Dixon Q-test or Grubbs test) or discarded immediately upon observation of the blunder.

---

3. Quantitative Statistical Metrics

Analyzing data sets requires computing measures of central tendency and dispersion:

Sample Mean ($\bar{x}$) and Median

• Sample Mean: The arithmetic average of all replicate observations ($N$): $\bar{x} = \frac{1}{N} \sum_{i=1}^N x_i = \frac{x_1 + x_2 + \dots + x_N}{N}$
• Sample Median: The middle observation when data are sorted in ascending rank order. For an even number of observations, the median is the arithmetic mean of the two central numbers. The median is resistant to extreme outliers.

Sample Standard Deviation ($s$)

The sample standard deviation quantifies the dispersion of data points around the mean: $s = \sqrt{\frac{\sum_{i=1}^N (x_i - \bar{x})^2}{N - 1}}$ The term $N - 1$ represents the degrees of freedom. Dividing by $N - 1$ rather than $N$ provides an unbiased estimate of the true population standard deviation ($\sigma$) from a finite sample size.

Relative Standard Deviation (RSD) / Coefficient of Variation (CV)

To compare precision between measurements expressed in different units or of vastly different magnitudes, standard deviation is normalized to the mean: $\text{RSD} = \left(\frac{s}{\bar{x}}\right) \times 100\%$ Lower RSD values indicate higher experimental precision (e.g., an RSD below $1.0\%$ reflects excellent analytical repeatability).

---

4. Quantitative Error Formulations

To express the deviation between an experimental outcome and the accepted literature standard:

Absolute Error ($E_{\text{abs}}$)

Absolute error communicates the absolute physical difference in the original measurement units: $E_{\text{abs}} = |x_{\text{experimental}} - x_{\text{accepted}}|$

Percent Error (Relative Error)

Percent error normalizes the discrepancy against the accepted literature value: $\text{Percent Error} = \frac{|x_{\text{experimental}} - x_{\text{accepted}}|}{x_{\text{accepted}}} \times 100\%$ In directional error analysis, signed percent error is used without absolute value bars: $\text{Signed \% Error} = \frac{x_{\text{experimental}} - x_{\text{accepted}}}{x_{\text{accepted}}} \times 100\%$ A positive signed percent error denotes overestimation; a negative signed error denotes underestimation.

---

5. Mathematical Propagation of Uncertainty

When calculated quantities derive from arithmetic combinations of several measured variables (each carrying an independent experimental uncertainty), uncertainties propagate according to specific mathematical rules.

Addition and Subtraction: Absolute Uncertainties in Quadrature

For a calculated quantity $z = x + y - w$, where each component has an independent absolute uncertainty ($\Delta x, \Delta y, \Delta w$): $\Delta z = \sqrt{(\Delta x)^2 + (\Delta y)^2 + (\Delta w)^2}$ Rule: For addition and subtraction, absolute uncertainties combine in quadrature.

Multiplication and Division: Relative Uncertainties in Quadrature

For a calculated quantity $z = \frac{x \cdot y}{w}$, fractional (or percentage) uncertainties combine in quadrature: $\frac{\Delta z}{z} = \sqrt{\left(\frac{\Delta x}{x}\right)^2 + \left(\frac{\Delta y}{y}\right)^2 + \left(\frac{\Delta w}{w}\right)^2}$ Multiplying both sides by $100\%$ gives: $\%\Delta z = \sqrt{(\%\Delta x)^2 + (\%\Delta y)^2 + (\%\Delta w)^2}$

Powers and Exponents

If a calculation involves a variable raised to a power, $z = x^n$: $\frac{\Delta z}{z} = |n| \left(\frac{\Delta x}{x}\right) \implies \%\Delta z = |n|(\%\Delta x)$ Notice that powers do not combine in quadrature; because the variable $x$ is correlated with itself, the relative uncertainty scales directly by the absolute exponent $|n|$.

---

6. Procedural Systematic Error Direction Analysis

Experimental questions often ask how a specific procedural mistake skews a calculated result. Trace the mathematical chain from experimental observation to final formula:

Case 1: Incomplete Drying in Gravimetric Analysis

• Scenario: A student synthesizes and isolates a barium sulfate ($\text{BaSO}_4$) precipitate to determine the mass percent of sulfate in an unknown fertilizer. The crucible is heated only once and weighed while residual moisture remains.
• Error Chain: Residual water adds unreacted mass $\implies m_{\text{precipitate}}$ recorded is falsely high $\implies$ calculated moles of $\text{SO}_4^{2-}$ are falsely high $\implies$ reported % $\text{SO}_4^{2-}$ in sample is falsely high.

Case 2: Air Bubble in Buret Tip During Titration

• Scenario: An unpurged air bubble resides in the buret tip below the stopcock prior to titrating an acid with standard $\text{NaOH}$. During delivery, the air bubble dislodges and fills with liquid titrant.
• Error Chain: Liquid fills the void space formerly occupied by air without reacting with the analyte $\implies$ final buret reading indicates a greater delivered volume than actually reacted $\implies V_{\text{titrant}}$ recorded is falsely high $\implies$ calculated moles of acid analyte are falsely high $\implies$ calculated molarity of unknown acid is falsely high.

Case 3: Dilution of Titrant by Wet Buret

• Scenario: A student rinses a buret with deionized water but forgets to rinse (condition) it with the standardized $\text{NaOH}$ titrant before filling.
• Error Chain: Residual water droplets dilute the titrant inside the buret $\implies$ actual titrant molarity is lower than labeled $\implies$ a larger volume of titrant is required to reach equivalence $\implies V_{\text{titrant}}$ measured is falsely high $\implies$ using the labeled titrant concentration in $n_{\text{analyte}} = M_{\text{labeled}} \times V_{\text{measured}}$ makes calculated analyte moles falsely high.

Case 4: Calorimeter Heat Loss to Surroundings

• Scenario: An exothermic neutralization reaction is carried out in an uninsulated coffee cup lacking a lid.
• Error Chain: Heat escapes rapidly to the surrounding room air $\implies$ maximum temperature reached is lower than theoretical adiabatic temperature $\implies \Delta T_{\text{solution}}$ measured is falsely low $\implies q_{\text{absorbed}} = m c_s \Delta T$ is falsely low $\implies$ magnitude of calculated $|\Delta H_{\text{neutralization}}|$ is falsely low (underestimated).

---

7. Worked Problem: Error Analysis in Molar Mass Determination

Problem: A student standardizes an unknown monoprotic weak acid by titrating a $0.2450\text{ g}$ sample with $0.1000\text{ M NaOH}$.

1. Trial 1 requires $24.80\text{ mL}$ of titrant.
2. Trial 2 requires $24.85\text{ mL}$ of titrant.
3. Trial 3 requires $24.75\text{ mL}$ of titrant. The accepted literature molar mass of the pure organic acid is $98.96\text{ g/mol}$.

Step 1: Compute Mean Titrant Volume and Standard Deviation $\bar{V} = \frac{24.80 + 24.85 + 24.75}{3} = 24.80\text{ mL} = 0.02480\text{ L}$ $\sum (V_i - \bar{V})^2 = (0.00)^2 + (+0.05)^2 + (-0.05)^2 = 0.0050\text{ mL}^2$ $s_V = \sqrt{\frac{0.0050}{3 - 1}} = \sqrt{0.0025} = 0.050\text{ mL}$ $\text{RSD} = \left(\frac{0.050}{24.80}\right) \times 100\% = 0.20\%\quad (\text{exceptional precision})$

Step 2: Calculate Experimental Molar Mass Since the acid is monoprotic ($\text{HA} + \text{NaOH} \to \text{NaA} + \text{H}_2\text{O}$), moles of acid equal moles of base: $n_{\text{acid}} = M_{\text{base}} \times \bar{V}_{\text{base}} = (0.1000\text{ mol/L})(0.02480\text{ L}) = 2.480 \times 10^{-3}\text{ mol}$ $\text{Molar Mass}_{\text{exp}} = \frac{0.2450\text{ g}}{2.480 \times 10^{-3}\text{ mol}} = 98.79\text{ g/mol}$

Step 3: Determine Absolute and Percent Error $E_{\text{abs}} = |98.79 - 98.96| = 0.17\text{ g/mol}$ $\text{Percent Error} = \frac{0.17\text{ g/mol}}{98.96\text{ g/mol}} \times 100\% = 0.17\%$

Step 4: Error Diagnostic Question If the student had spilled several drops of acid solution after weighing the sample but before titrating, what would be the effect on the calculated molar mass?

• Analysis: Fewer moles of acid would reside in the titration flask $\implies$ less $\text{NaOH}$ titrant required $\implies V_{\text{titrant}}$ would be falsely low $\implies$ calculated $n_{\text{acid}}$ would be falsely low $\implies$ dividing the original recorded sample mass ($0.2450\text{ g}$) by a smaller mole value would make the calculated molar mass falsely high.