Answer-first: Random Effects (RE) requires strong exogeneity because the RE estimator assumes that the unobserved unit-specific effect is uncorrelated with all regressors. If this assumption fails, RE becomes biased and inconsistent, making Fixed Effects the safer alternative.
Warm intro (and where to find the “why” pages)
If you’re staring at a Random Effects question at 11:47pm, feeling stuck, behind, or low‑key panicking because the RE vs. FE logic feels slippery, you’re not alone. RE questions look simple—“assume the effect is random”—but under exam pressure, students freeze when asked to explain why RE needs a stronger assumption than FE and what actually breaks when regressors correlate with unobserved heterogeneity.
If you’re rebuilding your foundation across topics, start at our Why Hub. If you need the full econometrics & time series roadmap for last-minute studying or troubleshooting, see Econometrics & Time Series.
Answer first
Random Effects requires strong exogeneity because it treats the unobserved unit effect as part of the error term. For RE to be unbiased, the regressors must be uncorrelated with both the idiosyncratic error and the unit-specific effect. If regressors correlate with the unobserved effect, RE inherits omitted variable bias.
Problem setup
Consider the panel model:
\[ y_{it} = \beta x_{it} + \alpha_i + u_{it}, \]
where:
- \(\alpha_i\) is the unobserved unit-specific effect,
- \(u_{it}\) is the idiosyncratic error.
RE assumes:
\[ \mathbb{E}[\alpha_i \mid x_{it}] = 0 \quad \text{for all } t. \]
This is the strong exogeneity condition.
Step-by-step solution (WordPress-safe MathJax)
Step 1: Understand the RE decomposition
RE treats the composite error as:
\[ \varepsilon_{it} = \alpha_i + u_{it}. \]
For RE to be consistent:
\[ \mathbb{E}[x_{it} \varepsilon_{it}] = 0. \]
This requires:
- \(\mathbb{E}[x_{it} \alpha_i] = 0\)
- \(\mathbb{E}[x_{it} u_{it}] = 0\)
Step 2: Why this is stronger than FE
FE allows \(\alpha_i\) to correlate with \(x_{it}\) because FE removes \(\alpha_i\) through demeaning. RE does not remove \(\alpha_i\); it keeps it in the error term.
Therefore, RE requires:
\[ \text{Cov}(x_{it}, \alpha_i) = 0. \]
Step 3: The RE estimator blends within and between variation
RE uses a quasi-demeaned transformation:
\[ y_{it} – \theta \bar{y}_i = \beta (x_{it} – \theta \bar{x}_i) + (u_{it} – \theta \bar{u}_i), \]
where:
\[ \theta = 1 – \sqrt{\frac{\sigma_u^2}{\sigma_u^2 + T \sigma_\alpha^2}}. \]
This mixes:
- within-unit variation (like FE), and
- between-unit variation (like pooled OLS).
If between-unit variation is biased, RE inherits that bias.
Step 4: State the identifying assumption clearly
RE requires:
\[ \mathbb{E}[\alpha_i \mid x_{i1}, \dots, x_{iT}] = 0. \]
This is stronger than FE’s strict exogeneity requirement.
Intuition
RE assumes that the unobserved unit effect is “noise,” not a confounder. If regressors correlate with that unobserved effect, RE breaks.
Think of RE as saying: “The differences between units are random.” But in real data, differences between units—ability, culture, geography—often correlate with regressors.
Common exam mistakes
- Claiming RE is more efficient than FE without checking assumptions.
- Forgetting that RE requires no correlation between regressors and \(\alpha_i\).
- Misinterpreting the Hausman test. It tests FE vs. RE consistency.
- Assuming RE is “more general.” It is actually more restrictive.
- Ignoring between-unit bias. RE blends within and between variation.
Why this matters
Random Effects is widely used in applied microeconomics, health economics, and panel data analysis. But it is only valid when regressors are uncorrelated with unobserved heterogeneity. Understanding this assumption helps you choose between FE and RE and interpret empirical results correctly.
Final summary
- RE requires no correlation between regressors and unobserved effects.
- RE blends within and between variation.
- FE is safer when unobserved heterogeneity correlates with regressors.
- The Hausman test helps decide between FE and RE.
Talk Directly to a Tutor, Not a Marketer
Call/Text: 510-398-0006
Email: tutor@californiagraduatetutor.com