Calculate the Sobel, Aroian or Goodman mediation test from path coefficients and standard errors, including the indirect effect, standard error, z score and two-sided p-value.

First-order variance estimate.

Use unstandardized coefficients and their positive standard errors from matching models and the same sample.

Independent variable → mediator.

Mediator → outcome, controlling for the independent variable.

Sobel Test Formula

The Sobel test is used in mediation analysis to evaluate whether an indirect effect is statistically different from zero. In a simple mediation model, an independent variable X affects a mediator M through the a path, and the mediator affects an outcome Y through the b path while controlling for X. The calculator offers the Sobel, Aroian and Goodman variance estimates for that indirect effect, each with a two-sided standard-normal p-value.

Indirect Effect = a b
SEab = √(b² SEₐ² + a² SEb²)
S = (a b) / (√(b² SEₐ² + a² SEb²))

The statistic S is commonly interpreted like a z-score. Larger absolute values indicate stronger evidence that the indirect effect is not zero under the normal approximation. The equations above show the Sobel variance. Aroian adds SE_a² SE_b² to that variance; Goodman subtracts it. The selected variance must be positive and numerically resolvable. These three conventional formulas use the same inputs. The two-sided probability is p = 2 Φ(−|z|), evaluated directly in the complementary tail; upper-tail and log probabilities remain useful for very small p-values.

Variable Meaning
a Effect of the independent variable on the mediator
b Effect of the mediator on the dependent variable while controlling for the independent variable
SEa Standard error of the a path coefficient
SEb Standard error of the b path coefficient
ab Estimated indirect effect
S Sobel test statistic for the indirect effect

How to Interpret the Sobel Test Statistic

Interpretation usually focuses on the absolute value of the statistic. The sign tells you the direction of the indirect effect, while the magnitude tells you how strongly the result differs from zero.

  • If a and b have the same sign, the indirect effect ab is positive.
  • If a and b have opposite signs, the indirect effect ab is negative.
  • If the standard errors are large, the Sobel statistic gets smaller, which makes significance less likely.
  • A result close to zero suggests weak evidence for mediation through the proposed mediator.
Two-Tailed Significance Level Approximate Critical Value Rule of Thumb
10% 1.645 If |S| > 1.645, the indirect effect is significant at the 0.10 level
5% 1.96 If |S| > 1.96, the indirect effect is significant at the 0.05 level
1% 2.576 If |S| > 2.576, the indirect effect is significant at the 0.01 level

Most mediation reporting uses the 5% two-tailed threshold, but your decision rule should match your study design and analysis plan. The calculator compares the unrounded p-value with 0.05; a rounded displayed value can equal 0.05 on either side of that boundary. Extremely small probabilities below the numerical range are shown as a bound with log10(p), not as an exact zero.

How to Use the Calculator

  1. Enter a, the coefficient from the model predicting the mediator from the independent variable.
  2. Enter b, the coefficient from the model predicting the outcome from the mediator while controlling for the independent variable.
  3. Enter SEa, the standard error associated with a.
  4. Enter SEb, the standard error associated with b.
  5. Choose the variance estimate, select Calculate, then review the indirect effect, estimated standard error, z statistic and two-sided p-value.

Use unstandardized coefficients and standard errors from the same sample, from matching regression outputs, and from the same model specification. Both standard errors must be strictly positive; coefficients can be signed or zero. If both coefficients are zero, Sobel has zero variance and no defined z statistic, Aroian gives z = 0 and p = 1, and Goodman is invalid. A zero indirect effect has zero direction. The conventional formulas do not accept an additional coefficient-covariance term.

Example

Suppose the estimated paths and standard errors are:

  • a = 0.5
  • b = 0.8
  • SEa = 0.2
  • SEb = 0.3
Indirect Effect = (0.5)(0.8) = 0.40
SEab = √((0.8²)(0.2²) + (0.5²)(0.3²)) = √(0.0481) ≈ 0.219
S = (0.40) / (0.219) ≈ 1.82

In this example, the indirect effect is positive and the Sobel two-sided p-value is approximately 0.0682. The statistic is slightly below the common 1.96 cutoff, so the effect is not statistically significant at the 0.05 level using the Sobel approach. This does not prove the absence of mediation.

When the Sobel Test Is Most Useful

  • When you already have regression coefficients and their standard errors
  • When you want a quick analytic test of mediation without resampling
  • When sample size is reasonably large and the normal approximation is more defensible
  • When you need a fast screening measure for an indirect pathway

Assumptions and Limitations

The Sobel test is convenient, but it is also known to be conservative in many real datasets. The main reason is that the product ab is often not perfectly normally distributed, especially in smaller samples.

  • Large-sample preference: The test performs better when the sample size is not small.
  • Normal approximation: It assumes the indirect effect can be adequately approximated with a normal test statistic.
  • Model dependence: Mis-specified regression models can make the result misleading.
  • Same-model requirement: The coefficient b must come from the outcome model that includes both the mediator and the independent variable.
  • Sensitivity to standard errors: Inflated standard errors can quickly reduce significance even when the indirect effect itself is meaningful.

In practice, researchers often pair the Sobel test with bootstrap confidence intervals because bootstrap methods can better handle the skewed distribution of indirect effects.

Common Input Mistakes

  • Using the wrong b coefficient from a model that does not control for the independent variable
  • Mixing coefficients from one model with standard errors from another
  • Using rounded values that are too coarse, which can noticeably change the final statistic
  • Confusing the sign of the indirect effect with statistical significance
  • Interpreting a non-significant Sobel result as proof that no mediation exists

Practical Interpretation Tips

  • Report the indirect effect ab along with the Sobel statistic.
  • Discuss both the direction and the strength of mediation.
  • If the result is borderline, consider whether sample size or standard error inflation may be limiting power.
  • If mediation is central to your analysis, consider supplementing the Sobel test with confidence intervals from resampling methods.