Week2 part2

A confidence interval provides a range of values that potentially contains a population parameter, such as a population proportion ( $p$ ).
The method for computing a $95\%$ confidence interval involves taking a random sample and applying a specific mathematical formula. This formula utilizes a critical value (sometimes referred to loosely as a "magical number") to determine the width of the interval.
The fundamental meaning of a $95\%$ confidence level is rooted in the long-run success rate of the methodology:
- If the process of taking a random sample and calculating a confidence interval is repeated many times (e.g., $100$ times), approximately $95\%$ of those resulting intervals are expected to capture the true population proportion.
- Conversely, approximately $5\%$ of those intervals (or $5$ out of $100$ ) will fail to capture the true parameter.
The confidence level is a statement about the reliability of the estimation process over time, not a specific probability assigned to a single, already-calculated interval.
For a single specific interval, it is unknown whether the true population parameter is contained within it. The parameter itself is a fixed value, whereas the intervals vary from sample to sample.

A common misunderstanding in interpreting results is the belief that a confidence interval describes the range where a sample mean will fall.
Correct Interpretation: A confidence interval for the mean is constructed to capture the population mean, not the sample mean. One cannot be $95\%$ sure that the mean of a future sample will fall within a previously calculated $95\%$ confidence interval.
Calculation Example: In a scenario where a $99\%$ confidence interval is calculated over $1,000$ trials, the underlying method is mathematically expected to capture the population parameter $990$ times ( $0.99 \times 1,000 = 990$ ).

Descriptive Statistics: This involves reporting the measures of center, specifically the mean and median. The choice between reporting the mean or median depends on the nature of the data. These should be accompanied by measures of spread and an analysis of the data's shape.
Parameter Estimation: This involves using sample data to create confidence intervals, which provide an estimate for unknown population parameters.
Hypothesis Testing: This process uses sample data to decide between two competing claims (hypotheses) about a population. It tests the validity of a specific claim rather than just providing an estimate.

Research Question: The investigation centers on whether Parkinson's disease can be detected by sense of smell.
Background: Joy Milne (a Scottish woman) noticed a distinct change in the smell of her husband, Les, prior to his diagnosis with Parkinson's disease. This led to the hypothesis that the disease produces a unique olfactory signature.
Significance of Detection by Smell:
- Early Diagnosis: Identifying the disease earlier allows for the commencement of treatment at an earlier stage, which generally results in better patient outcomes.
- Cost-Effectiveness: Smell-based detection could potentially be less expensive than existing diagnostic technologies.
- Research Impact: Scientific proof of a specific scent would shift the perspective on Parkinson's research and treatment development.
Initial Experimental Test: Researchers provided Joy with $12$ T-shirts to smell. Some were worn by Parkinson's patients and some by healthy controls. She was asked to identify which shirts belonged to the patients.

Competing Hypotheses: Hypothesis testing evaluates two opposing claims, known as the null and alternative hypotheses.
Null Hypothesis ( $H_0$ ): This is the baseline claim that there is no effect or no difference. In the context of Joy Milne's test, $H_{0}$ states that she cannot smell Parkinson's and is simply guessing.
- Mathematically, if she is guessing between two options (Parkinson's or no Parkinson's), the proportion ( $p$ ) she would get correct is $p = 0.5$ .
Alternative Hypothesis ( $H_a$ or $H_1$ ): This is the claim the researcher seeks to support with evidence. In this case, $H_a$ states that Joy can detect the disease at a rate better than random chance.
- Mathematically, this is expressed as p > 0.5.
Testing Logic: The test determines how unlikely the observed data (the experimental result) would be if the null hypothesis were true.

Simulation of Random Guessing: To quantify how likely success is by sheer luck, an experiment can be simulated using cards (which have no smell). If a person guesses on $12$ cards, they are essentially replicating the behavior of a "guesser" in the T-shirt experiment.
Simulation Results for $12$ Trials:
- Getting exactly $6$ out of $12$ correct is highly likely, occurring in nearly $30\%$ of experiments where someone is just guessing.
- Getting $12$ out of $12$ correct is rare; simulation shows this occurs only about $4\%$ of the time by chance.
Defining the P-Value: The P-value is the probability of obtaining a sample statistic that is as extreme or more extreme than the observed sample result, assuming the null hypothesis ( $H_0$ ) is true.
Simulation-Based P-Values:
- The P-value for achieving $12$ out of $12$ correct identifications is $0.04$ ( $4\%$ ).
- The P-value for achieving $11$ or more out of $12$ correct identifications is $0.06$ ( $6\%$ ), calculated by adding the probabilities of getting exactly $11$ and exactly $12$ .

Question: If someone is purely guessing and cannot smell the disease, how many out of $12$ shirts would they be expected to identify correctly?
Answer: On average, a person guessing would get $6$ correct due to the $50/50$ probability of each choice.
Question: How many correct identifications are required to convince a scientist that the ability is real?
Answer: There is no single correct answer; it depends on the tolerance for error. For high-stakes medical diagnoses (like cancer), a high degree of certainty (e.g., $100\%$ ) is often required. For less critical claims, a scientist might be satisfied with $10$ or $11$ out of $12$ .
Question: What is specifically wrong with the statement "I am $95\%$ sure that the mean of a sample will fall within a $95\%$ confidence interval for the mean"?
Answer: The statement is incorrect because the interval is designed to estimate the population mean, not to predict where the sample mean will land. The sample mean is what was used to build the interval in the first place.