Lecture 7 – Confidence Intervals, Z vs T, and Minitab Demonstration

Session Overview

• Two main tools used: Minitab (two blank scales) and Lecture Seven slides (PowerPoint).

• Hands-on project: generate population data, then a random sample from that population, compute descriptive statistics, and form a confidence interval for a population mean. A volunteer will come up to perform the steps on the teacher’s computer; the class observes and then analyzes the results.

• Emphasis on understanding both manual calculations and software-assisted calculations (Minitab), and on understanding the underlying concepts (confidence intervals, z vs t, degrees of freedom, etc.).

Generating a Large Population Data Set (Population)

• Use Minitab Calc -> Random Data -> Normal Distribution to generate population data.

  • Distribution: Normal distribution.

  • Number of values: e.g., 10,000 (large population).

  • Store in a column (e.g., Column C1).

  • Parameters to fix for the class demo (example): mean $\mu = 205$, standard deviation $\sigma = 16$.

  • Note: The exact values used by the instructor are not known to the students during the exercise; the volunteer inputs their own $\mu$ and $\sigma$, and the rest of the class uses those same values for comparability.

• Rationale: This big dataset represents the population from which a sample will be drawn; it allows a controlled demonstration of population vs sample statistics.

• Conceptual point: In practice, we often do not know the true population parameters ($\mu$, $\sigma$); the project demonstrates estimation from a sample.

Generating a Random Sample from the Population

• In Minitab, Calc -> Random Data -> Sample from Columns.

• Setup:

  • Source: select the population column (e.g., C1).

  • Sample size: choose a number larger than 30 to leverage the CLT (e.g., 36).

  • Destination: put the new sample in a separate column (e.g., C3) to visually separate population and sample.

  • Sampling option: with replacement (the teacher notes that sampling with vs without replacement doesn’t matter much for this exercise; replacement keeps the process simple).

• After generating the sample, compute descriptive statistics for the sample (mean and standard deviation) using Minitab (Stat -> Basic Statistics -> Display Descriptive Statistics for the sampled column(s)).

• Class activity: Students are asked to report the mean and standard deviation of the sample; the instructor uses these to illustrate estimation and the construction of confidence intervals.

Descriptive Statistics and Key Quantities

• Population statistics (theoretical, often unknown in practice):

  • Mean: $\mu$ (unknown in real life; here we pretend to know it only to illustrate the process).

  • Standard deviation: $\sigma$ (unknown in real life; we may estimate with $s$ from the sample).

• Sample statistics (calculated from the drawn sample):

  • Sample size: $n$.

  • Sample mean: $\bar{x}$.

  • Sample standard deviation: $s$.

• Practical note: In the demonstration, the volunteer provides the true population mean and standard deviation (for teaching purposes) so students can compare the sample-based estimates with known values. In real-world practice, these population parameters are unknown.

Confidence Intervals: Core Idea and Notation

• Goal: Provide a range that plausibly contains the population mean $\mu$ with a given confidence level.

• General formula (when sigma is known):

  • $\bar{x} \pm z_{\alpha/2} \cdot \frac{\sigma}{\sqrt{n}}$

  • Here, $z_{\alpha/2}$ is the standard normal critical value for the two tails corresponding to the desired confidence level.

• Common shorthand values:

  • For 95% confidence, $z{\alpha/2} = z{0.025} \approx 1.96$.

  • For 68% confidence, $z_{\alpha/2} \approx 1$ (by empirical rule).

  • For 99.7% confidence, $z_{\alpha/2} \approx 3$ (by empirical rule).

• Practical note: The empirical rule (68-95-99.7) provides approximate guidance; exact values come from the standard normal distribution.

• When sigma is unknown (the common real-world case): replace $\sigma$ with the sample standard deviation $s$ and use the $t$-distribution with $df = n - 1$:

  • $\bar{x} \pm t_{\alpha/2,\, df}\cdot \frac{s}{\sqrt{n}},\quad df = n-1$

• Degrees of freedom: $df = n - 1$.

• Interpretation nuance: The phrase “95% confidence” means that, in repeated sampling, 95% of the constructed intervals would contain the true $\mu$; for a single interval, there is still a 5% chance that the interval does not contain $\mu$.

• When $n$ grows large, the $t$-distribution approaches the standard normal (i.e., $t{\alpha/2, df} \to z{\alpha/2}$ as $df \to \infty$).

Z vs T: Practical Distinctions

• Known sigma: Use the z-distribution (standard normal). Commonly used value is $z_{\alpha/2}$.

• Unknown sigma: Use the t-distribution with $df = n - 1$; the t-distribution is more spread out for small $n$ and converges to the normal as $n$ grows.

• Intuition: When $\sigma$ is unknown, estimating it from the data adds extra variability; the t-distribution accounts for this extra variability via its heavier tails.

• Visual intuition: As degrees of freedom increase (larger $n$), the t-curve becomes more like the standard normal curve.

Notation and Concepts Explained

• $\alpha$ (significance level) is defined as $\alpha = 1 - \text{confidence level}$.

• The critical value $z{\alpha/2}$ (or $t{\alpha/2, df}$) corresponds to the point where the total tail probability equals $\alpha/2$ in each tail.

• Why split alpha by 2? Because the confidence interval is symmetric around $\bar{x}$ and we allocate half of the tail probability to each side of the distribution.

• If you want a nonstandard confidence level (e.g., 84%), you can find the corresponding $z$ or $t$ value by inspecting the distribution (or using software). For 84% in the middle, the tails together sum to 16%, so each tail is 8%; the corresponding $z$ value is approximately $z \approx 1.405$.

• The instructor also demonstrates that for 87% confidence, the corresponding $t$ value (with a given df) can be found using the t-distribution plot and is about $t \approx 1.538$ for $df = 54$ (example numbers from the demo).

Example: 84% Confidence Interval (Illustrative)

• Suppose we want a central 84% interval (unusual but used for demonstration):

  • Tail probability each side: $(1 - 0.84)/2 = 0.08$.

  • For the standard normal, the corresponding critical value is approximately $z_{0.08} \,\text{two-tailed} \approx 1.405$.

• If using known $\sigma$ and a sample mean $\bar{x}$ with sample size $n$:

  • Interval: $\bar{x} \pm 1.405 \cdot \frac{\sigma}{\sqrt{n}}.$

• If unknown $\sigma$, replace with $s$ and use $t_{0.08, df}$ accordingly.

Example: 87% Confidence Interval with Unknown Sigma (Manual vs Minitab)

• Given a sample size $n = 55$, sample mean $\bar{x} = 612.46$, sample standard deviation $s = 47.12$:

  • Standard error: $SE = \frac{s}{\sqrt{n}} = \frac{47.12}{\sqrt{55}} \approx 6.383.$

  • For 87% confidence, the two-tailed significance is $\alpha = 1 - 0.87 = 0.13$, so each tail has probability $\alpha/2 = 0.065$.

  • The corresponding $t$ value is about $t_{0.065, df=54} \approx 1.538$ (from the t-distribution plot or table).

  • Margin of error: $ME = t_{0.065, 54} \cdot SE \approx 1.538 \cdot 6.383 \approx 9.83.$

  • Confidence interval: $\left( \bar{x} - ME , \; \bar{x} + ME \right) \approx (602.63, 622.29).$

• Validation with Minitab (one-sample t CI): Using the same $n$, $\bar{x}$, and $s$ yields a very similar interval (numbers may differ slightly due to rounding); the manual calculation and Minitab should agree closely.

• Takeaway: Software (Minitab) can compute the exact t-interval; manual computation helps you understand the components and verify correctness.

Practical Notes on Confidence Intervals and Interpretation

• When you have $\sigma$ known, the interval is centered at $\bar{x}$ with width determined by $\sigma$ and $n$ via the normal distribution.

• When $\sigma$ is unknown, the interval uses $s$ and the t-distribution, with the width depending on the sample size (through $df = n-1$).

• In practice, you often report both the interval and the method (z vs t) to reflect whether $\sigma$ was known or estimated.

• Special values to memorize (as per class instruction):

  • Common 95% interval uses $z{\alpha/2} = 1.96$ (exactly, $z{0.025} \approx 1.959964$).

  • The empirical rule provides quick approximations: roughly $\pm 1\sigma$ for 68%, $\pm 2\sigma$ for 95%, $\pm 3\sigma$ for 99.7% around the mean.

  • A precise 95% interval with unknown $\sigma$ uses $t_{0.025, df}$ instead of 1.96, and the value depends on $df$.

Notes on Alpha, Z, and Confidence Levels

• $\alpha$ (significance) is the complement of the confidence level: $\alpha = 1 - \text{confidence}$.

• $z_{\alpha/2}$ is the critical value for the standard normal distribution with half of the significance in each tail.

• The notation $z{\alpha/2}$ is a conventional way to express the cutoffs; in practice you may see $z{0.025}$ for a 95% CI.

• For nonstandard confidence levels (e.g., 83%, 87%), you can use software (or normal/t distribution tables) to find the corresponding critical value.

• For paper-based exams, the instructor notes that you may be allowed to use approximate rules (68-95-99.7) or the exact values such as 1.96 for 95%; for computer-based exams, you can rely on software values.

Degrees of Freedom and the Shape of the t-Distribution

• Degree of freedom (df) is defined as $df = n - 1$ for a single sample.

• The t-distribution depends on $df$:

  • With $df = 1$ (small sample size, e.g., $n=2$), the t-curve is very flat and with heavy tails relative to the normal.

  • As $df$ increases, the t-curve becomes more like the standard normal; by $df \approx 30$, it is very close to the normal distribution.

• This is why, for small samples, the t-interval is wider than the z-interval (to account for extra uncertainty from estimating $\sigma$ with $s$).

• In teaching demonstrations, the instructor often uses $df = n-1$ to illustrate how the interval width changes with sample size.

Manual Calculation vs Software (Minitab) Validation

• Steps to compute a confidence interval manually (illustrative):

  • Compute $\bar{x}$ from the sample.

  • Compute $s$ (sample standard deviation).

  • Choose the desired confidence level (e.g., 95%, 87%, 84%); determine the appropriate critical value ($t{\alpha/2, df}$ or $z{\alpha/2}$).

  • Compute the standard error: $SE = \frac{s}{\sqrt{n}}$.

  • Compute the margin of error: $ME = critical_value \cdot SE$.

  • Form the interval: $\left( \bar{x} - ME, \; \bar{x} + ME \right).$

• Software validation (Minitab): Use

  • For known $\sigma$: Stat -> Basic Statistics -> 1-Sample Z

  • For unknown $\sigma$: Stat -> Basic Statistics -> 1-Sample T

  • In both cases, provide the data (or summary: $n$, $\bar{x}$, $s$ or $\sigma$) and set the desired confidence level (default 95%).

  • Compare the interval produced by Minitab to the manual calculation to check for rounding differences.

• The instructor also demonstrates how to compute the t critical value manually using a t-distribution plot (Graph -> Probability Distribution -> View Probability) for nonstandard confidence levels (e.g., 84% in the middle) by setting df and tail probabilities. This reinforces understanding of what the critical value represents.

Special Topics Discussed During the Demo

• When population parameters are unknown in practice, we rely on sample-based estimates and adjust the distribution accordingly (switch from z to t).

• The CLT underpins using the normal approximation for many sampling distributions, but when sigma is unknown and the sample is not large, the t-distribution is more appropriate.

• The demonstration covers both manual computation (to understand each piece of the interval) and software-based computation (to perform accurately and efficiently).

• The instructor emphasizes not blindly trusting software; students should be able to validate the results manually and understand the underlying mechanics.

Practical Advice for Exam and Quiz Preparation

• Common exam defaults: when not specified, you may be asked to compute a 95% confidence interval.

• If the exam is paper-based, you may be expected to recall standard critical values (e.g., 1.96 for 95% CI) or to use elementary z/t rules; for computer-based questions, you may be asked to use software to obtain the exact interval.

• For 95% CI with unknown sigma, you should use the t-distribution with df = n - 1; the exact critical value depends on df and the chosen confidence level.

• Memorize two key numbers mentioned in class notes: a practical 30 as a rule-of-thumb for when z and t are very close (n around 30 or larger) and 1.96 as the exact z critical value for 95% confidence.

• If you need to compute a nonstandard confidence level manually, you can do so by locating the appropriate z or t critical value using distribution plots or tables and then apply the standard formula.

Quick Summary of Core Formulas

• Known sigma (normal):

  • $\text{CI} = \bar{x} \pm z_{\alpha/2} \cdot \frac{\sigma}{\sqrt{n}}.$

• Unknown sigma (t-distribution):

  • $\text{CI} = \bar{x} \pm t_{\alpha/2,\, df} \cdot \frac{s}{\sqrt{n}},\quad df = n-1.$

• Empirical rule (approximate):

  • $P(|X - \mu| \le \sigma) \approx 0.68,$

  • $P(|X - \mu| \le 2\sigma) \approx 0.95,$

  • $P(|X - \mu| \le 3\sigma) \approx 0.997.$

Reminder: Exam Logistics Mentioned in the Session

• Quiz Three will cover chapters 5 and 6 (not 7); lecture seven content may be incomplete for the quiz.

• There is no ACAP check on this weekend’s material; the professor will resume AC/AP checks next week.

• Students are encouraged to schedule meetings for help; the instructor offers hours and mentions office hours to clarify concepts and calculations.

• VT-7 (Confidence Interval topic) and related topics will continue; students should ensure submission of their work (including 87% CI example) and practice manual verification alongside Minitab results.

Key Takeaways for the Exam

• Know when to use z vs t and how to interpret which distribution is appropriate based on knowledge of sigma and the sample size.

• Be able to compute and interpret a confidence interval for a population mean, including the meaning of the interval and the confidence level.

• Understand degrees of freedom and how they affect the critical value in t-distributions.

• Be comfortable performing manual calculations and validating them with software outputs.

• Be able to extract and use summary statistics ($n$, $\bar{x}$, $s$) to form a confidence interval, including the effect of nonstandard confidence levels by adjusting the critical value.

End of Notes

• If you want, I can convert these notes into a condensed cheat-sheet with the most essential formulas and a quick step-by-step workflow for both manual and Minitab-based confidence interval construction.

It sounds like you're looking for a conceptual framework that explains the 'why' and 'how' behind confidence intervals, focusing on the relationships between components and their practical implications. Let's break down the process and key elements to help you visualize and feel how everything connects and computes, filling the gap of estimating an unknown population mean.

1. The Core Problem and Goal: The 'Gap' to Fill

• Problem: In statistics, we almost never know the true population mean ($\mu$) or population standard deviation ($\sigma$). We can't measure everyone or everything.

• Goal: The primary goal is to estimate this unknown population mean ($\mu$) meaningfully. Instead of a single guess (a 'point estimate'), which is almost certainly wrong, we want a range of plausible values for $\mu$. This range is our confidence interval (CI).

• How it fills the gap: The CI gives us a calculated range ($(\bar{x} - ME, \bar{x} + ME)$) that, with a certain confidence, we expect to contain the true population mean. It's a way to acknowledge the uncertainty inherent in using a sample to understand a population.

2. Bucketing the Components: What's What?

• Population (The 'True' but Unknown World):

  • Population Mean ($\mu$): The number we want to find. It's fixed but unknown.

  • Population Standard Deviation ($\sigma$): Represents the spread of data in the entire population. Fixed but usually unknown.

• Sample (Our 'Window' into the Population):

  • Sample Size ($n$): How many observations we collect. This is crucial for precision.

  • Sample Mean ($\bar{x}$): The average ($\Sigma x / n$) of our collected data. This is our best single guess for $\mu$.

  • Sample Standard Deviation ($s$): The spread of our collected data. This is our best estimate for $\sigma$.

• Confidence Level (Our 'Certainty'):

  • Usually expressed as a percentage (e.g., 90%, 95%, 99%). It's not the probability that OUR specific interval contains $\mu$. Instead, it refers to the long-run proportion of intervals that would contain $\mu$ if we repeated the sampling process many times. It influences the width of our interval.

• Alpha ($\alpha$):

  • Significance level, $\alpha = 1 - \text{Confidence Level}$. It's the total probability of error in hypothesis testing, but for CIs, it's the probability that our interval does not contain the true mean $\mu$. We split it into two tails ($\alpha/2$).

3. Understanding Relationships and Patterns: Z vs. T

This is a critical distinction that dictates which formula and critical value to use:

• The Decision Flow:

  1. Is $\sigma$ (population standard deviation) known?

     • YES (Rare, but possible in specific scenarios): Use the Z-distribution (standard normal). This implies we have a very precise understanding of the population's spread. The critical value ($z_{\alpha/2}$) comes from the standard normal curve.

     • NO (Very Common in Real-World): Use the T-distribution. When $\sigma$ is unknown, we have to estimate it using the sample standard deviation ($s$). This adds an extra layer of uncertainty.

       • Degrees of Freedom ($df$): The t-distribution needs this. For a single sample mean, $df = n - 1$. This accounts for the fact that we're estimating $\sigma$ from the sample, making our estimate of $\mu$ less certain.

       • Shape: The t-distribution is flatter and has 'heavier tails' than the Z-distribution, especially for small $df$ (small $n$). This means it makes our confidence interval wider, reflecting that increased uncertainty.

• Why does this matter? Using 's' instead of '$\sigma$' makes our estimate inherently more variable. The t-distribution is designed to 'punish' us for this extra uncertainty by spreading out the critical values, leading to wider (more conservative) intervals, especially for small sample sizes.

4. Visualizing Motion and Dynamics: How Changes Affect the CI

Imagine the confidence interval as a 'net' we're casting to catch $\mu$. How does our 'net' change?

• Impact of Sample Size ($n$):

  • Larger $n$ (more data): The standard error ($SE = s/\sqrt{n}$ or $\sigma/\sqrt{n}$) decreases, making the interval narrower (a tighter net). We become more precise. $df$ also increases, making the t-distribution closer to the Z-distribution.

  • Smaller $n$ (less data): The standard error increases, making the interval wider (a looser net). We are less precise. $df$ decreases, making the t-distribution spread out more, further widening the interval.

• Impact of Confidence Level (e.g., 90% to 99%):

  • Higher Confidence (e.g., 99%): We want to be more 'certain' our interval contains $\mu$. To achieve this, we need a wider interval (a bigger net). This means the critical value ($z<em>{\alpha/2}$ or $t</em>{\alpha/2, df}$) will increase.

  • Lower Confidence (e.g., 90%): We are willing to be less 'certain.' This allows for a narrower interval (a smaller net). The critical value will decrease.

• Impact of Variability ($s$ or $\sigma$):

  • Larger Standard Deviation: The data points are more spread out. This naturally leads to a wider interval, as there's more inherent variability to account for. Standard error increases.

  • Smaller Standard Deviation: The data points are clustered more tightly. This leads to a narrower interval, as there's less inherent variability. Standard error decreases.

5. Function, Purpose, and Computation: The Formula in Action

The overall function of the CI formula is to calculate the Margin of Error (ME) and add/subtract it from the sample mean to get the plausible range.

• General Formula: $\text{CI} = \bar{x} \pm \text{Margin of Error (ME)}$

• Purpose of Each Formula Part:

  • $\bar{x}$: This is our point estimate, the center of our interval.

  • The $\pm$ sign: Indicates that the interval extends equally above and below the sample mean.

  • Critical Value ($z<em>{\alpha/2}$ or $t</em>{\alpha/2, df}$): This value is about 'how many standard errors away from the mean' we need to go to achieve our desired confidence. It's a multiplier for our uncertainty.

  • Standard Error ($SE = s/\sqrt{n}$ or $\sigma/\sqrt{n}$): This is the estimated standard deviation of the sampling distribution of the mean. It tells us how much we expect sample means to vary from the true population mean due to random sampling. It's the 'unit' of uncertainty.

• Computation Flow:

  1. Collect Sample Data: Get $n$, calculate $\bar{x}$ and $s$. (These are your 'inputs').

  2. Determine Known vs. Unknown Population $\sigma$: Decide whether to use Z or T.

  3. Choose Confidence Level: This determines $\alpha$ and $\alpha/2$. (e.g., 95% CI -> $\alpha=0.05$ -> $\alpha/2=0.025$).

  4. Find Critical Value: Look up $z<em>{\alpha/2}$ or $t</em>{\alpha/2, df}$ (where $df = n-1$) from tables or software, based on your decision in step 2 and confidence level in step 3.

  5. Calculate Standard Error (SE): Divide $s$ (or $\sigma$ if known) by $\sqrt{n}$. This quantifies the variability of sample means.

  6. Calculate Margin of Error (ME): Multiply the Critical Value (from step 4) by the SE (from step 5). This is the 'half-width' of your interval.

  7. Construct Interval: Add and subtract the ME from $\bar{x}$. $( \bar{x} - ME, \bar{x} + ME )$.

By understanding these relationships, patterns, and dynamic movements, you can not only compute the interval but also intuitively grasp what each part represents and how changes in your data or desired certainty affect the final estimate. It's about building a robust 'net' to capture the unknown population mean.