Internal Validity

Internal Validity

  • Internal validity assesses the quality of the experimental design in supporting a causal claim.
  • It focuses on whether the experiment was well-designed to avoid confounds.
  • A well-designed experiment should lead to only one possible claim.
  • A confound arises when there are alternative explanations for the results, undermining the claim.
  • Internal validity contrasts with external validity.
  • External validity examines whether the experimental results generalize to real-world settings.
  • Internal validity focuses solely on the design of the experiment itself.
  • It asks: Does the experiment unambiguously support the researcher's claim, or are there other potential explanations?
  • Good internal validity means the research design effectively rules out alternative explanations, supporting the claim.

Example: Preschool Quality and College Admissions

  • The claim: "Better preschools lead to better colleges."
  • Original claim was an association claim: "People who go to better preschools tend to get into better colleges."
  • The revised claim is a causal claim, asserting that better preschools cause better college admissions.
  • The causal claim posits that a good educational environment in preschool positively affects later college prospects.
  • Private preschools can cost $30,000-$40,000 per year and require consideration if the investment is warranted.
  • If the causal claim is true, investing in high-quality preschool is critical for long-term educational success.
  • This suggests that the early years are crucial for building a foundation that extends to college.

X and Y Variables

  • Using X and Y variables to illustrate the causal claim:
    • X = Better preschool
    • Y = Better college
  • The claim: Better preschool (X) causes better college (Y).
  • The question: Is it this simple, or are there other factors involved?
  • Taking for granted there's a relationship: children who go to better preschools eventually tend to get into better colleges.
  • Is preschool the only reason for those advantages?
  • Association vs. causation: The association claim may be correct, but not necessarily the causal claim.

Potential Confounding Factors

  • Money (Socioeconomic Status)
    • Money can be a confound. Families who can afford better preschools might also afford better colleges.
    • This suggests that socioeconomic status, rather than preschool quality, is the primary driver.
    • More exclusive, private, and resource-rich institutions are generally considered “better.”
  • Parental Investment in Education
    • The level of parental involvement in a child's education also plays a significant role.
    • Parents who invest in better preschools may also consistently support and push their children academically.
    • Parental involvement throughout a child's education could be the reason for college success rather than preschool quality alone.
  • These two variables are not mutually exclusive.
  • Teasing out if it's really the preschool and not these other variables requires complex statistical analysis.

Diagramming the Confound

  • Revisiting the X and Y relationship with potential confounds:
    • Better preschool (X) causes better college (Y)? - Questionable.
    • Does higher socioeconomic status (Z) cause better college (Y)?
    • Does greater parental investment in education (Z) cause better college (Y)?
  • Z represents a third variable affecting the relationship between X and Y.
  • It undermines the claim that preschool quality alone leads to better college outcomes.

The Third Variable Problem

  • The third variable, Z, gets in the way of establishing a clear causal relationship between X and Y.
  • It introduces a confound, confusing the true cause of better college admissions.
  • We cannot definitively say that preschool quality is the determining factor.
  • Understanding confounds is critical for evaluating research claims.
  • If preschool isn't the key factor but money, it may not matter as much which preschool the family chooses.

Definition of Confound

  • A third variable is a confound.
  • Confound undermines our ability to interpret the validity of this statement: Better preschool causes better college.
  • A confound results in multiple potential causes: is it X or Z that is causing Y?
  • Third Variable Problem: X \rightarrow Y , but also Z \rightarrow Y.
  • The possibility of a third variable means we are confused as to which is the cause.
  • Recognizing confounds is essential for critical thinking and evaluating research.
  • Skeptical inquiry helps identify these confounding factors.

Ruling Out Confounding Factors

  • Good internal validity means potential confounds have been successfully ruled out.
  • An experiment's design eliminates alternative explanations which makes the causal claim more believable.
  • Demonstrating internal validity involves designing experiments to rule out alternative factors.
  • Ruling out confounds is how to find evidence that the claim that better preschools lead to better colleges is internally valid.
  • It requires controlling for socioeconomic status, parental investment, and other potential influences.
  • This is more complex than demonstrating an association claim.
  • To control socioeconomic status:
    • Recruit families across different socioeconomic levels.
    • Track their progress from preschool to college.
    • Use statistical techniques to determine the unique impact of preschool quality, controlling for income.
  • To measure parental investment:
    • Quantify parental involvement (e.g., hours spent on schoolwork).
    • Measure the impact on college admissions.
  • These studies are challenging and time-consuming.
  • Causal claims require substantial evidence to rule out alternative explanations.

Challenges in Establishing Causation

  • Longitudinal studies are complex; families' circumstances can change over time.
  • Attrition (participants dropping out) can introduce bias.
  • Unaccounted variables at home or other situations can affect academic outcomes.
  • Establishing causal relationships in real-world settings is inherently messy.
  • Association claims easier to support, showing correlation without causation.
  • Causal claims demand more robust evidence to support them.
  • Next is how the four validities can be used to carefully evaluate the there different categories of claims.