Statistical Inference & Estimation: Concepts, Methods & Applications

### **TAKEAWAYS:**
- **Estimation vs. Inference**: Point estimation (single value) vs. interval estimation (range with confidence) to quantify uncertainty in population parameters.
- **Confidence Intervals**: Constructed using sample statistics (e.g., $\bar{x}\pm z_{\alpha /2}\frac{\sigma }{\sqrt{n}}$) to estimate population means/proportions with a specified confidence level (e.g., 95% or 99%).
- **Hypothesis Testing Framework**: Null ($H_0$) and alternative ($H_1$) hypotheses, test statistics (e.g., $t$-test, $z$-test), and decision rules based on critical values or $p$-values.
- **Sample Size Determination**: Calculating $n$ for desired precision (margin of error) and confidence, critical for survey design and quality control.
- **Applications**: Real-world problems like comparing IT officer productivity (CVS study), toothpaste quality control, or voter preference polls.
- **Key Formulas**: Confidence intervals for means ($\sigma$ known/unknown), proportions, and sample size formulas for means/proportions.

---

### **1. Introduction to Statistical Inference**
Statistical inference is the process of drawing conclusions about a **population** based on **sample data**. It includes:
- **Estimation**: Using sample statistics to estimate population parameters (e.g., $\bar{x}$ for $\mu$).
- **Hypothesis Testing**: Making decisions about population parameters using sample evidence.

#### **Key Definitions**
| Term                     | Definition                                                                                     |
|--------------------------|-------------------------------------------------------------------------------------------------|
| **Parameter**            | A numerical characteristic of a population (e.g., $\mu$, ${\sigma }^2$, $p$).               |
| **Statistic**            | A numerical characteristic of a sample (e.g., $\bar{x}$, $s^2$, $\hat{p}$).               |
| **Point Estimation**     | Using a single value (statistic) to estimate a parameter (e.g., $\bar{x}$ for $\mu$).       |
| **Interval Estimation**  | Providing a range (confidence interval) where the parameter likely lies (e.g., 95% CI).       |
| **Confidence Level**     | Probability ($1-\alpha$) that the confidence interval contains the true parameter.           |
| **Margin of Error (ME)** | Half the width of the confidence interval: $ME=z_{\alpha /2}\cdot \text{SE}$, where SE is the standard error. |

---

### **2. Point Estimation**
Point estimation provides a single value as an estimate of a population parameter. Common estimators:
- **Mean**: $\hat{\mu }=\bar{x}$
- **Proportion**: $\hat{p}=\frac{X}{n}$, where $X$ is the number of successes in a sample of size $n$.
- **Variance**: ${\hat{\sigma }}^2=s^2=\frac{\sum (x_i-\bar{x}{)}^2}{n-1}$

#### **Properties of Good Estimators**
1. **Unbiasedness**: $E(\hat{\theta })=\theta$ (e.g., $\bar{x}$ is unbiased for $\mu$).
2. **Consistency**: $\hat{\theta }$ converges to $\theta$ as $n\to \infty$.
3. **Efficiency**: Lowest variance among unbiased estimators (e.g., $\bar{x}$ is efficient for $\mu$ if $\sigma$ is known).

#### **Example: Estimating Mean Lifespan of Computers**
Given the dataset for Dell, Lenovo, and HP computers:
| Serial | Brand  | Lifetime (years) |
|--------|--------|-------------------|
| 1      | Dell   | 15                |
| 2      | Lenovo| 10                |
| 3      | HP     | 9                 |
| 4      | Dell   | 12                |
| ...    | ...    | ...               |

**Step 1**: Calculate the sample mean ($\bar{x}$) for each brand.
**Step 2**: Use $\bar{x}$ as the point estimate for the population mean ($\mu$) of each brand.

---
### **3. Interval Estimation**
Interval estimation provides a range (confidence interval) for a population parameter, accounting for sampling variability.

#### **Confidence Interval for Population Mean ($\mu$)**
##### **Case 1: $\sigma$ Known (Z-Interval)**
$$\bar{x}\pm z_{\alpha /2}\cdot \frac{\sigma }{\sqrt{n}}$$
- $z_{\alpha /2}$: Critical value from the standard normal distribution (e.g., $z_{0.025}=1.96$ for 95% CI).
- **Example**: A sample of 36 IT officers has a mean working hours $\bar{x}=8$ with $\sigma =1.5$. Find the 95% CI for $\mu$.
  $$8\pm 1.96\cdot \frac{1.5}{\sqrt{36}}=8\pm 0.49⟹(7.51,8.49)$$

##### **Case 2: $\sigma$ Unknown (T-Interval)**
$$\bar{x}\pm t_{\alpha /2,n-1}\cdot \frac{s}{\sqrt{n}}$$
- $t_{\alpha /2,n-1}$: Critical value from the $t$-distribution with $n-1$ degrees of freedom.
- **Example**: For $n=10$, $\bar{x}=8$, $s=1.5$, and 95% CI:
  $$t_{0.025,9}\approx 2.262⟹8\pm 2.262\cdot \frac{1.5}{\sqrt{10}}=8\pm 1.07⟹(6.93,9.07)$$

#### **Confidence Interval for Population Proportion ($p$)**
$$\hat{p}\pm z_{\alpha /2}\cdot \sqrt{\frac{\hat{p}(1-\hat{p})}{n}}$$
- **Example**: In a poll of 1000 voters, 58% favor a candidate. Find the 95% CI for $p$.
  $$0.58\pm 1.96\cdot \sqrt{\frac{0.58\cdot 0.42}{1000}}=0.58\pm 0.031⟹(0.549,0.611)$$

#### **Confidence Level vs. Margin of Error**
| Confidence Level | $z_{\alpha /2}$ | Margin of Error (for $n=1000$, $\hat{p}=0.5$) |
|------------------|-------------------|----------------------------------------------------|
| 90%              | 1.645             | $1.645\cdot \sqrt{0.25/1000}=0.026$            |
| 95%              | 1.96              | $0.031$                                         |
| 99%              | 2.576             | $0.041$                                         |

---
### **4. Hypothesis Testing**
Hypothesis testing evaluates claims about population parameters using sample data.

#### **Steps in Hypothesis Testing**
1. **State Hypotheses**:
   - Null hypothesis ($H_0$): Default assumption (e.g., $\mu ={\mu }_0$).
   - Alternative hypothesis ($H_1$): Claim to test (e.g., $\mu \ne {\mu }_0$, $\mu >{\mu }_0$).
2. **Choose Significance Level ($\alpha$)**: Commonly 0.05 or 0.01.
3. **Calculate Test Statistic**:
   - For means: $t=\frac{\bar{x}-{\mu }_0}{s/\sqrt{n}}$ (if $\sigma$ unknown).
   - For proportions: $z=\frac{\hat{p}-p_0}{\sqrt{p_0(1-p_0)/n}}$.
4. **Determine Critical Value or $p$-value**:
   - Compare test statistic to critical value (e.g., $t_{0.025,n-1}$).
   - Or find $p$-value (probability of observing test statistic under $H_0$).
5. **Make Decision**:
   - Reject $H_0$ if test statistic falls in rejection region or $p$-value < $\alpha$.
   - Fail to reject $H_0$ otherwise.

#### **Example: Testing Mean Working Hours**
**Claim**: IT officers work an average of 8 hours/day ($\mu =8$).
**Sample**: $n=36$, $\bar{x}=8.5$, $s=1.5$, $\alpha =0.05$.
**Hypotheses**:
- $H_0:\mu =8$
- $H_1:\mu \ne 8$ (two-tailed test).

**Test Statistic**:
$$t=\frac{8.5-8}{1.5/\sqrt{36}}=\frac{0.5}{0.25}=2$$

**Critical Value**: $t_{0.025,35}\approx 2.03$ (from $t$-table).
**Decision**: Since $2<2.03$, fail to reject $H_0$. No significant evidence that $\mu \ne 8$.

#### **Comparison: Z-Test vs. T-Test**
| Feature          | Z-Test                          | T-Test                          |
|------------------|---------------------------------|---------------------------------|
| **When to Use**  | $\sigma$ known or $n\ge 30$ | $\sigma$ unknown, $n<30$  |
| **Distribution** | Standard normal ($Z$)         | $t$-distribution              |
| **Degrees of Freedom** | N/A                     | $n-1$                         |
| **Example**      | Large sample sizes             | Small samples (e.g., $n=10$)   |

---

### **5. Sample Size Determination**
Sample size ($n$) is calculated to ensure the estimate is precise (small margin of error) with a given confidence level.

#### **For Population Mean ($\mu$)**
$$n={\left(\frac{z_{\alpha /2}\cdot \sigma }{ME}\right)}^2$$
- **Example**: To estimate $\mu$ with $ME=0.5$, $\sigma =2$, and 95% confidence:
  $$n={\left(\frac{1.96\cdot 2}{0.5}\right)}^2=61.47⟹n=62$$

#### **For Population Proportion ($p$)**
$$n=\frac{z_{\alpha /2}^2\cdot \hat{p}(1-\hat{p})}{M{E}^2}$$
- If $\hat{p}$ is unknown, use $\hat{p}=0.5$ (worst-case scenario).
- **Example**: For $ME=0.05$, 95% confidence:
  $$n=\frac{1.96^2\cdot 0.5\cdot 0.5}{0.05^2}=384.16⟹n=385$$

#### **Adjustments**
- For finite populations, use $n_{adj}=\frac{n}{1+\frac{n-1}{N}}$, where $N$ is population size.
- For pilot studies, use initial $\hat{p}$ or $\bar{x}$.

---

### **6. Applications in Real-World Problems**
#### **Example 1: Toothpaste Quality Control**
- **Problem**: 20 out of 400 tubes leak. Estimate the proportion of defective tubes with 96% confidence.
- **Solution**:
  - $\hat{p}=\frac{20}{400}=0.05$.
  - $z_{0.02}=2.054$ (for 96% CI).
  - $ME=2.054\cdot \sqrt{\frac{0.05\cdot 0.95}{400}}=0.024$.
  - **CI**: $0.05\pm 0.024⟹(0.026,0.074)$.

#### **Example 2: Comparing Two Means (Independent T-Test)**
**Problem**: Test if the mean lifetime of Dell (${\bar{x}}_1=13.5$, $s_1=2$, $n_1=5$) and HP (${\bar{x}}_2=9.5$, $s_2=1.5$, $n_2=5$) computers are significantly different ($\alpha =0.05$).

**Hypotheses**:
- $H_0:{\mu }_1={\mu }_2$
- $H_1:{\mu }_1\ne {\mu }_2$

**Test Statistic** (assuming unequal variances):
$$t=\frac{{\bar{x}}_1-{\bar{x}}_2}{\sqrt{\frac{s_1^2}{n_1}+\frac{s_2^2}{n_2}}}=\frac{13.5-9.5}{\sqrt{\frac{4}{5}+\frac{2.25}{5}}}=\frac{4}{\sqrt{1.25}}=3.54$$

**Critical Value**: $t_{0.025,8}\approx 2.306$ (df = $n_1+n_2-2=8$).
**Decision**: Reject $H_0$ ($3.54>2.306$). Significant difference in lifetimes.

#### **Example 3: Confidence Interval for Difference of Means**
Using the same data:
$$({\bar{x}}_1-{\bar{x}}_2)\pm t_{\alpha /2}\cdot \sqrt{\frac{s_1^2}{n_1}+\frac{s_2^2}{n_2}}$$
$$(13.5-9.5)\pm 2.306\cdot \sqrt{1.25}=4\pm 2.68⟹(1.32,6.68)$$
**Interpretation**: We are 95% confident that the true difference in lifetimes is between 1.32 and 6.68 years.

---

### **7. Common Pitfalls and Misconceptions**
1. **Confidence Interval ≠ Probability**: A 95% CI means there’s a 95% chance the interval method produces a range containing $\mu$, not that $\mu$ has a 95% chance of falling in the interval.
2. **Overlapping CIs ≠ No Difference**: Overlapping CIs do not guarantee no significant difference (depends on sample size and variability).
3. **P-Value ≠ Probability of $H_0$**: The $p$-value is the probability of observing data as extreme as the sample, assuming $H_0$ is true—not the probability that $H_0$ is true.
4. **Ignoring Assumptions**: T-tests assume normality (for small $n$) and equal variances (for two-sample tests). Violations can lead to incorrect conclusions.

---

### **8. Exam Tips**
#### **What Examiners Look For**
1. **Correct Hypotheses**: Always state $H_0$ and $H_1$ clearly.
2. **Assumptions**: Mention assumptions (e.g., normality, independence) before calculations.
3. **Step-by-Step Calculations**: Show all steps, especially for test statistics and CIs.
4. **Interpretation**: Explain the practical meaning of results (e.g., "We reject $H_0$ at 5% significance, suggesting...").
5. **Units and Clarity**: Include units (e.g., "years," "hours") and label all terms.

#### **Common Exam Questions and How to Approach Them**
| Question Type                          | Key Steps                                                                                     |
|----------------------------------------|-----------------------------------------------------------------------------------------------|
| **Confidence Interval for Mean**       | Identify $\sigma$ known/unknown → use Z or T formula. Plug in values carefully.              |
| **Hypothesis Test for Mean**           | State hypotheses, calculate test statistic, compare to critical value or find $p$-value.    |
| **Sample Size Calculation**            | Use the correct formula for mean/proportion. Show substitution of values.                     |
| **Comparing Two Means (T-Test)**       | Check independence, equal variances (use Welch’s t-test if unequal). Calculate pooled SE.     |
| **Proportion Estimation**              | Use $\hat{p}=\frac{X}{n}$ and the proportion CI formula.                                |
| **Interpretation**                     | Relate results to the context (e.g., "The data suggests Dell computers last longer...").     |

#### **Marking Scheme Insights**
- **Definitions (2-3 marks)**: Clearly define terms like "confidence level," "point estimate," or "test statistic."
- **Calculations (5-7 marks)**: Partial credit for correct formulas, even if final answer is wrong.
- **Interpretation (2-3 marks)**: Always explain what the result means in the context of the problem.
- **Assumptions (1-2 marks)**: Mention normality, independence, or equal variance where applicable.

#### **Practice Focus Areas**
1. **Confidence Intervals**: Master both Z and T intervals for means and proportions.
2. **Hypothesis Testing**: Practice one-tailed and two-tailed tests with real-world data.
3. **Sample Size**: Know when to use the mean or proportion formula and how to adjust for finite populations.
4. **Comparing Groups**: Be comfortable with independent t-tests and interpreting CIs for differences.

---
### **9. Worked Example: Comprehensive Problem**
**Problem**:
A study was conducted among IT officers in Kathmandu Valley to quantify the effect of age and working hours per day on Computer Vision Syndrome (CVS). A random sample of 40 officers was taken, with the following data on daily working hours ($X$) and CVS severity score ($Y$):

| Officer | Age (years) | Working Hours (X) | CVS Score (Y) |
|---------|-------------|-------------------|---------------|
| 1       | 28          | 8                 | 5             |
| 2       | 35          | 9                 | 7             |
| ...     | ...         | ...               | ...           |
| 40      | 50          | 7                 | 4             |

**Tasks**:
1. Construct a 95% confidence interval for the mean working hours (${\mu }_X$) if the sample mean $\bar{x}=8.5$ hours and sample standard deviation $s=1.2$.
2. Test at 5% significance whether the mean CVS score (${\mu }_Y$) is greater than 5, given $\bar{y}=6$ and $s_y=1.5$.
3. Determine the sample size needed to estimate the mean working hours with a margin of error of 0.5 hours and 95% confidence, assuming $\sigma =1.2$.

---
#### **Solution**:
**1. Confidence Interval for ${\mu }_X$**:
- **Given**: $\bar{x}=8.5$, $s=1.2$, $n=40$, $\alpha =0.05$.
- **Formula**: $\bar{x}\pm t_{\alpha /2,n-1}\cdot \frac{s}{\sqrt{n}}$.
- **Critical Value**: $t_{0.025,39}\approx 2.023$ (from $t$-table).
- **Calculation**:
  $$ME=2.023\cdot \frac{1.2}{\sqrt{40}}=2.023\cdot 0.1897\approx 0.383$$
  $$CI=8.5\pm 0.383⟹(8.117,8.883)$$
- **Interpretation**: We are 95% confident that the true mean working hours for IT officers is between 8.12 and 8.88 hours.

**2. Hypothesis Test for ${\mu }_Y$**:
- **Hypotheses**:
  - $H_0:{\mu }_Y\le 5$
  - $H_1:{\mu }_Y>5$ (one-tailed test).
- **Test Statistic**:
  $$t=\frac{\bar{y}-{\mu }_0}{s/\sqrt{n}}=\frac{6-5}{1.5/\sqrt{40}}=\frac{1}{0.237}\approx 4.22$$
- **Critical Value**: $t_{0.05,39}\approx 1.685$ (one-tailed).
- **Decision**: Since $4.22>1.685$, reject $H_0$.
- **Conclusion**: There is significant evidence at 5% level to conclude that the mean CVS score is greater than 5.

**3. Sample Size Calculation**:
- **Given**: $ME=0.5$, $\sigma =1.2$, $\alpha =0.05$ ($z_{0.025}=1.96$).
- **Formula**:
  $$n={\left(\frac{z_{\alpha /2}\cdot \sigma }{ME}\right)}^2={\left(\frac{1.96\cdot 1.2}{0.5}\right)}^2=(4.704{)}^2\approx 22.13$$
- **Result**: $n=23$ (round up to ensure precision).

---
### **10. Summary Table of Key Formulas**
| Scenario                          | Formula                                                                                     | Notes                                  |
|-----------------------------------|---------------------------------------------------------------------------------------------|----------------------------------------|
| **CI for $\mu$ ($\sigma$ known)** | $\bar{x}\pm z_{\alpha /2}\cdot \frac{\sigma }{\sqrt{n}}$                              | Use Z-table.                          |
| **CI for $\mu$ ($\sigma$ unknown)** | $\bar{x}\pm t_{\alpha /2,n-1}\cdot \frac{s}{\sqrt{n}}$                              | Use T-table.                          |
| **CI for Proportion**             | $\hat{p}\pm z_{\alpha /2}\cdot \sqrt{\frac{\hat{p}(1-\hat{p})}{n}}$                     | Ensure $np\ge 10$ and $n(1-p)\ge 10$. |
| **Hypothesis Test for $\mu$**   | $t=\frac{\bar{x}-{\mu }_0}{s/\sqrt{n}}$                                                  | Degrees of freedom: $n-1$.          |
| **Hypothesis Test for $p$**     | $z=\frac{\hat{p}-p_0}{\sqrt{p_0(1-p_0)/n}}$                                          | Use normal approximation.             |
| **Sample Size for $\mu$**       | $n={\left(\frac{z_{\alpha /2}\cdot \sigma }{ME}\right)}^2$                          | Adjust for finite populations.        |
| **Sample Size for $p$**         | $n=\frac{z_{\alpha /2}^2\cdot \hat{p}(1-\hat{p})}{M{E}^2}$                             | Use $\hat{p}=0.5$ if unknown.     |

---
### **11. Additional Resources**
- **Books**:
  - *Statistical Methods for the Social Sciences* by Alan Agresti and Barbara Finlay.
  - *Introduction to the Practice of Statistics* by Moore, McCabe, and Craig.
- **Software**:
  - Use R/Python for calculations:
    ```python
    # Confidence Interval in Python
    from scipy import stats
    stats.t.interval(0.95, 39, loc=8.5, scale=1.2/np.sqrt(40))
    ```
  - Excel: `=T.INV.2T(0.05, 39)` for critical $t$-value.
- **Online Tools**:
  - [Stat Trek](https://stattrek.com/) for calculators and tutorials.
  - [VassarStats](http://vassarstats.net/) for hypothesis testing.

---
### **12. Common Mistakes to Avoid**
1. **Incorrect Degrees of Freedom**: For two-sample t-tests, use $n_1+n_2-2$ (not $n_1+n_2$).
2. **Wrong Distribution**: Use $t$-distribution for small samples with unknown $\sigma$, not $Z$.
3. **Ignoring Normality**: For non-normal data, consider non-parametric tests (covered in Unit 4).
4. **Misinterpreting $p$-Values**: A $p$-value of 0.05 does not mean "5% chance $H_0$ is true."
5. **Rounding Errors**: Carry intermediate steps to 3-4 decimal places to avoid rounding mistakes.

---
### **Exam Tip**
- **For TU Exams**:
  - **Theory Questions**: Focus on definitions (e.g., "What is a confidence interval?") and comparisons (e.g., "Z-test vs. T-test").
  - **Numerical Problems**: Always show calculations step-by-step. Partial marks are given for correct formulas and intermediate steps.
  - **Interpretation**: Examiners emphasize practical interpretation. For example, if testing CVS scores, state whether the result suggests a health concern or not.
- **For PU Exams**:
  - Expect more applied problems (e.g., sample size determination for surveys).
  - Be ready to justify your choice of test (e.g., "I used a t-test because the sample size is small and $\sigma$ is unknown").
- **NEB Exams**:
  - Emphasize clarity and correctness over speed. NEB often tests conceptual understanding alongside calculations.
- **General Advice**:
  - **Practice Past Papers**: TU/PU/NEB often repeat question types (e.g., CI for proportions, two-sample t-tests).
  - **Time Management**: Allocate 1.5–2 minutes per mark. For example, a 10-mark question should take ~15–20 minutes.
  - **Check Units**: Ensure answers are in the correct units (e.g., hours, years, percentages).