Analytical Chemistry Study Notes: Normal Distribution, Hypothesis Testing, F-Test, and t-Test

Standard Deviation, Degrees of Freedom, and the Normal Distribution Curve

• Degree of Freedom ($df$):

  • Defined as the total number of measurements or tests ($n$) minus 1 ($n - 1$) when calculating the sample standard deviation ($s$).

  • Formula: $df = n - 1$

  • Crucial parameter required to consult statistical tables ($F$-table, $t$-table).

• The Standard Normal Distribution ($z$-Score):

  • Standardized distance metric ($z$) indicating how many standard deviations ($s$ or $\text{σ}$) a measured value ($x$) lies away from the sample mean ($\bar{x}$ or $\text{μ}$).

  • Formula for $z$: $z = \frac{x - \bar{x}}{s}$

  • Absolute deviation formula: $z = \frac{|x - \bar{x}|}{s}$

  • Random errors cause individual experimental measurements to fluctuate around the mean, yielding values that are randomly larger or smaller than the average.

• Symmetry and Properties of the Bell Curve:

  • The bell-shaped normal distribution curve is completely symmetric around the central mean value ($z = 0$).

  • The area to the left of the mean ($-\infty$ to $\bar{x}$) and the area to the right of the mean ($\bar{x}$ to $+\infty$) are exact $50\%/50\%$ ($0.500/0.500$) mirror images of each other.

  • The midpoint ($z = 0$) represents zero deviation between the measured value and the sample average ($x = \bar{x}$).

  • Moving outward from the center represents values progressively further from the average, expressed as increments of standard deviation ($1\sigma$, $2\sigma$, $3\sigma$).

• Area Under the Curve as Probability:

  • The probability of measuring a value within a specified range is equal to the area under the normal curve over that range.

  • Total area under the curve from negative infinity ($-\infty$) to positive infinity ($+\infty$) equals $1.00$ (or $100\%$).

  • Area from the central mean ($z = 0$) to $z = 1$: $0.341$ or $34.1\%$.

  • Area within one standard deviation ($\pm 1\sigma$, from $z = -1$ to $z = +1$): $34.1\% \times 2 = 68.3\%$ ($0.683$).

  • Area within two standard deviations ($\pm 2\sigma$, from $z = -2$ to $z = +2$): $95.5\%$ ($0.955$).

  • Tail area calculation from $z = 1$ to $+\infty$: Calculated by subtracting the central-to-one-sigma area from the total half-curve area ($50\% - 34.1\% = 15.9\%$, or approximately $16\%$).

Practical Applications of z-Scores and Tail Probabilities

• Coin Toss Experiment Setup:

  • Experimental trial consisting of tossing a set of $50\,\text{coins}$ simultaneously.

  • Theoretical probability predicts a mean ($\bar{x}$) of $25\,\text{heads}$.

  • Calculated standard deviation ($s$) for the set is $3.54$.

  • Range of outcomes per toss: Minimum of $0\,\text{heads}$ to maximum of $50\,\text{heads}$.

  • The overall experiment involves tossing the $50\,\text{coins}$ for a total of $400\,\text{times}$.

• Calculating Probability for 15 or Fewer Heads ($x \le 15$):

  • Difference from mean: $|15 - 25| = 10$

  • $z$-score calculation: $z = \frac{|15 - 25|}{3.54} = \frac{10}{3.54} = 2.8$

  • Standard normal table lookup for $z = 2.8$ gives an area from the mean to $z = 2.8$ of $0.4974$ (or $49.74\%$ / $0.497$).

  • Left-tail area ($x \le 15$): Calculated by subtracting the table value from the entire left half of the distribution ($50\%$ or $0.5000$):     $\text{Area} = 0.5000 - 0.4974 = 0.0026$

  • Percentage representation: $50\% - 49.7\% = 0.3\%$ (or exact table value $0.26\%$).

  • Expected frequency out of $400\,\text{tosses}$: $400 \times 0.0026 = 1.04\,\text{tosses}$.

• Calculating Probability for 32 or More Heads ($x \ge 32$):

  • Absolute difference from mean: $32 - 25 = 7$

  • $z$-score calculation: $z = \frac{32 - 25}{3.54} = \frac{7}{3.54} = 1.977 \approx 2$

  • Standard normal table lookup for $z = 2$ yields an area from the mean to $z = 2$ of $0.4772$ (or $47.7\%$ / $0.447$).

  • Right-tail area ($x \ge 32$): Subtracting table area from right-half total ($50\%$):     $\text{Area} = 50\% - 47.72\% = 2.28\% \approx 2.3\%$

• Additional Reference z-Score Calculations:

  • For a mean difference of $0.18$ and standard deviation of $0.89$:     $z = \frac{0.18}{0.89} = 0.2$

  • Table area from mean to $z = 0.2$ is $0.079$ ($7.9\%$).

  • For $z = 1.5$:

  • Table area from mean to $z = 1.5$ is $0.4332$ ($43.32\%$).

Hypothesis Testing and Variance Comparison (The F-Test)

• The Null Hypothesis ($H_0$):

  • Assumes that two sample sets or measurement methods are derived from the exact same population.

  • Assumes that observed differences between means ($\bar{x}_1, \bar{x}_2$) or standard deviations ($s_1, s_2$) are caused strictly by random errors, not systematic errors.

  • Criterion for rejection: Reject the null hypothesis if there is less than a $5\%$ probability ($p < 0.05$) that the observed differences between the samples are due solely to random variation.

• Distinction Between F-Test and t-Test:

  • $F$-test: Compares the variance ($s^2$) or standard deviation similarity of two analytical methods to evaluate relative precision.

  • $t$-test: Compares the mean values ($\bar{x}$) of two data sets to evaluate accuracy and statistical identity.

• $F$-Test Formula and Setup:

  • Variance is defined as the square of the sample standard deviation ($s^2$).

  • Formula for calculated $F$ value ($F_{\text{calc}}$):     $F_{\text{calc}} = \frac{s_1^2}{s_2^2}$

  • Ordering rule: Place the larger sample standard deviation (variance) in the numerator ($s_1^2$) and the smaller sample standard deviation (variance) in the denominator ($s_2^2$).

  • Requirement: $F_{\text{calc}}$ must always be greater than or equal to 1 ($F_{\text{calc}} \ge 1$).

• Evaluating the $F$-Test:

  • Numerator degrees of freedom: $df_1 = n_1 - 1$ (corresponding to the larger standard deviation dataset).

  • Denominator degrees of freedom: $df_2 = n_2 - 1$ (corresponding to the smaller standard deviation dataset).

  • Locate $F_{\text{table}}$ at the intersection of $df_1$ and $df_2$ in the critical value table at the $95\%$ confidence level ($\alpha = 0.05$).

  • Decision Rule:

  • If $F_{\text{calc}} > F_{\text{table}}$: Reject the null hypothesis; standard deviations are significantly different.

  • If $F_{\text{calc}} \le F_{\text{table}}$: Accept the null hypothesis; standard deviations are not significantly different (differences are consistent with random variation).

Case Study: Instrument Validation Using the F-Test

• Scenario Context:

  • Monitoring bicarbonate anion ($\text{HCO}_3^-$) levels in post-race testing to enforce prohibitions against administration of buffering agents used to neutralize lactic acid accumulated during activity.

  • Authorities must validate a new instrument by comparing its precision against an established original instrument.

• Experimental Data:

  • Original instrument: Mean $\bar{x}_2 = 36$, standard deviation $s_2 = 0.28$, sample size $n_2 = 10$.

  • New (substitute) instrument: Mean $\bar{x}_1 = 34.2$, standard deviation $s_1 = 0.47$, sample size n_1 = 4$.\n\n* **Degrees of Freedom:**\n * Numerator (s_1 = 0.47$):$df_1 = n_1 - 1 = 4 - 1 = 3\n * Denominator (s_2 = 0.28$):$df_2 = n_2 - 1 = 10 - 1 = 9\n\n* **F-Value Calculation:**\n  F_{\text{calc}} = \frac{(0.47)^2}{(0.28)^2} = \frac{0.2209}{0.0784} = 2.82\n\n* **Table Lookup and Decision:**\n * Critical value from table at df_1 = 3$and$df_2 = 9$($95\%$confidence level):$F_{\text{table}} = 5.08\n * Comparison: F_{\text{calc}} (2.82) < F_{\text{table}} (5.08)\n * Conclusion: Accept the null hypothesis. The standard deviations s_1$and$s_2 are not significantly different. The new instrument possesses comparable precision to the original instrument and can be safely adopted.\n\n# Confidence Intervals and Expressing Measurement Uncertainty\n\n* **Three Standard Methods for Expressing Uncertainty:**\n 1. Reporting result as mean plus or minus standard deviation: \bar{x} \pm s\n 2. Comparing experimental mean (\bar{x}$) to a known accepted true value ($\text{μ}) using confidence intervals.\n 3. Comparing two sets of experimental data (means \bar{x}1$and$\bar{x}_2) to evaluate mutual consistency.\n\n* **Confidence Interval Formula:**\n  \text{CI} = \bar{x} \pm \frac{t \times s}{\sqrt{n}}\n * Where \bar{x}$is sample mean,$t$is Student's$t$table value at$df = n - 1$,$s$is standard deviation, and$n is total number of measurements.\n\n* **Carbohydrate Content Case Study (n = 5):**\n * Degrees of freedom: df = 5 - 1 = 4\n * **50\% Confidence Interval Calculation:**\n * Table Student's t$value at$df = 4$($50\%$level):$t{50\%} = 0.741\n * Interval equation: \text{CI}{50\%} = \bar{x} \pm \frac{0.741 \times s}{\sqrt{5}}\n * Meaning: Repeated measurements have a 50\% probability of capturing the true population mean within this interval.\n * **90\% Confidence Interval Calculation:**\n * Table Student's t$value at$df = 4$($90\%$level):$t{90\%} = 2.132\n * Interval equation: \text{CI}{90\%} = \bar{x} \pm \frac{2.132 \times s}{\sqrt{5}}\n * Meaning: Repeated measurements have a 90\% probability of capturing the true population mean within this expanded interval.\n\n* **Comparing Experimental Results to Known True Values (\text{μ}):**\n * Calculate the 95\%$confidence interval around the experimental sample mean ($\bar{x}).\n * If a known reference/true value (e.g., 1570$) falls inside the calculated$95\% confidence interval, the experimental result is not statistically different from the known value.\n * Blood Sugar Control Case Study: Evaluating patient family blood sugar measurements against a control sample standard reference value of 99$at the$95\% confidence level to check for significant systematic deviation.\n\n# Comparison of Means Using Pooled Standard Deviation (t-Test)\n\n* **Prerequisite Condition:**\n * Before performing a two-sample t$-test to compare means, an$F$-test must first confirm that the standard deviations of the two datasets are not significantly different ($F{\text{calc}} \le F_{\text{table}}).\n\n* **Pooled Standard Deviation (s_{\text{pooled}}):**\n * Combines standard deviations of both sample sets into a weighted variance standard.\n * Formula:\n    s_{\text{pooled}} = \sqrt{\frac{s_1^2(n_1 - 1) + s_2^2(n_2 - 1)}{n_1 + n_2 - 2}}\n\n* **Calculated t$-Value ($t_{\text{calc}}):**\n * Formula:\n    t_{\text{calc}} = \frac{|\bar{x}1 - \bar{x}_2|}{s{ ext{pooled}}} \sqrt{\frac{n_1 n_2}{n_1 + n_2}}\n\n* **Decision Rule for Two-Sample t-Test:**\n * Total degrees of freedom: df = n_1 + n_2 - 2\n * If t_{\text{calc}} > t_{\text{table}}$at$95\% confidence level: Reject null hypothesis; the two experimental means are significantly different.\n * If t_{\text{calc}} \le t_{\text{table}}: Accept null hypothesis; the two experimental means are consistent within expected random error limits.\n\n# Questions & Discussion\n\n* **Degrees of Freedom Selection for Tables:**\n * Student query regarding selecting the appropriate degrees of freedom when table values do not match standard increments (e.g., selecting between df = 3$or$df = 4).\n * Explanation: Degrees of freedom depend strictly on the mathematical formula applied (df = n - 1$for single variance,$df = n_1 + n_2 - 2 for pooled two-sample comparison).\n\n* **Manual vs. Software Standard Deviation Calculations in Examinations:**\n * Student query regarding whether standard deviation must be calculated manually step-by-step during examinations.\n * Clarification: Manual standard deviation computation for full datasets takes approximately 5\,\text{minutes}$$. In examination settings, raw standard deviations or software computations (e.g., Excel output) are provided to streamline analytical evaluations.