Business Econometrics Notes

Business Econometrics (BUS-G350)

Kelley School of Business - Indiana University

Lecture 5 Outline
  • Hypothesis Testing

  • Scientific Method Definition and Details

  • Research Paper: Predictive Analytics for Business


Hypothesis Testing

  • Aim: Assess hypothesis about population

    • Example Hypothesis:

    • Population mean expenditure: πœ‡ = $105

    • Hypothesized mean: πœ‡β‚• = $105

  • Data Collection: Collect sample and calculate the sample mean (𝑋̅).

    • Suppose 𝑋̅ = $107.

  • Question: Does πœ‡ = $105 still hold?


Formal Hypothesis Test Procedure
  1. Collect a random sample.

  2. Compute the sample mean (𝑋̅).

  3. Compare sample mean (𝑋̅) to hypothesized mean (πœ‡β‚•).

  4. Decision Rule:

    • Reject the hypothesis if it is β€œtoo far away” from the sample mean.

    • Otherwise, β€œfail to reject” the hypothesis.


Determining "Too Far Away"
  • Key Question: How far is β€œtoo far away”?

    • Need further information to make a determination.


Central Limit Theorem (CLT)
  • Key Insight:

    • If sufficient data is present, then the sampling distribution of the sample mean is approximately normally distributed:
      XΜ… hicksim N(πœ‡, rac{𝜎²}{N})

  • This insight allows us to create a test statistic.

  • Specifically, we will use the t-statistic.


t-Statistic Formula
  • Definition: t = rac{XΜ… - ΞΌβ‚•}{S_{XΜ…}}

    • Where:

    • 𝑋̅: Sample Mean

    • 𝑆_{XΜ…}: Standard Error of the Sample Mean

    • 𝑆: Sample Standard Deviation

    • 𝑁: Sample Size

    • πœ‡β‚•: Hypothesized population mean


t-Statistic Analysis
  • Comparison to Standardization:

    • Z<em>i=racX</em>iβˆ’XΛ‰SZ<em>{i} = rac{X</em>{i} - \bar{X}}{S}

  • When N is large, by the CLT we have
    XΜ… hicksim N(πœ‡, rac{𝜎²}{N})


Distribution of t-Statistic Under the Null Hypothesis
  • If the null hypothesis is true: t = rac{XΜ… - πœ‡}{S/ ext{sqrt}(N)}

    • If N is large, then:

    • thicksimN(0,1)t hicksim N(0, 1)


Deductive Logic of Hypothesis Testing
Premises:
  1. If the hypothesis is true.

  2. If N > 30.

Conclusion:
  • Then
    thicksimN(0,1)t hicksim N(0, 1)

Steps in Hypothesis Testing
  1. Compute t-statistic.

  2. Determine if t-statistic appears to come from the standard normal distribution:

    • If yes, consistent with hypothesis, fail to reject.

    • If no, inconsistent with hypothesis, reject the hypothesis.


Understanding "Too Far Away"
  • If the hypothesis holds, then:

  • thicksimN(0,1)t hicksim N(0, 1)

  • This allows for the calculation of p-values.


p-Value Definition
  • Definition: The probability of obtaining a t-statistic as extreme as the one observed.

  • p-Value Calculation Steps:

    1. Compute t-statistic (t).

    2. Place t and -t on standard normal x-axis.

    3. Calculate probability > t and < -t.

    4. Summate to find p-value.


Hypothesis Testing Example
  • Given:

    • πœ‡β‚• = 105,

    • 𝑋̅ = 107,

    • 𝑁 = 10,000,

    • 𝑆 = 1.

  • Example t-statistic calculation:
    t=rac107βˆ’1051/extsqrt(10000)=2ext,thust=200.t = rac{107 - 105}{1/ ext{sqrt}(10000)} = 2 ext{, thus } t = 200.


Interpretation of the p-Value
  • t-stat: 200.

  • p-value:

    • Difficult to calculate manually; traditionally done via β€œtable lookup.”

    • Current methods utilize statistical software to automate calculations.

    • p-value < 0.001

    • Interpretation: Hypothesis is likely false; reject hypothesis.


Cut-off Values for p-Values
  • Establish significance levels for hypothesis testing:

    • Reject hypothesis if p < 10% (10% significance).

    • Reject hypothesis if p < 5% (5% significance).

    • Reject hypothesis if p < 1% (1% significance).


Rejection Region
  • The graphical representation of rejection regions in hypothesis testing denoting where the null hypothesis can be rejected based on significant t-values.


Hypothesis Testing Errors
  • Type I Error: Rejecting the null hypothesis when it is true.

  • Type II Error: Failing to reject the null when it is false.

    • Both capture misinterpretations of conclusions.


Scientific Method Overview
  • Interest in causal relationships: Does A cause B?

    • If affirmative, degree of causation?

  • Process:

    1. Formulate a hypothesis (e.g., β€œThis drug reduces cancer risk.”).

    2. Conduct an experiment in a controlled environment.

    • Experiment: A structured investigation to test hypotheses.

    • Treatment: Something administered to at least one participating group.

    • Treatment Effect: Change in outcome as a result of treatment.


Conclusion Drawing from Data
  • Analysis Steps:

    • Compare outcomes between treated and untreated groups.

    • Typically, build confidence intervals for the treatment effects and/or conduct hypothesis tests.


Average Treatment Effect (ATE)
  • Objective: Find causal effect of treatment (T).

    • Assign participants randomly to either:

    • Treatment group (𝑇𝑖 = 1)

    • Control group (𝑇𝑖 = 0).

  • Formula for ATE:
    ATE=YΛ‰<em>Tβˆ’YΛ‰</em>UATE = \bar{Y}<em>{T} - \bar{Y}</em>{U}