Malmquist Bias

Malmquist bias is a foundational observational selection effect in stellar astronomy, extragalactic distance-scale calibration, and observational cosmology in which a flux-limited (apparent-magnitude-limited) survey preferentially samples the high-luminosity tail of any astronomical population that possesses non-zero intrinsic scatter in luminosity (σ_M > 0). First derived by Swedish astronomer K. Gunnar Malmquist in 1920 and 1922 at Lund Observatory, the bias arises from three-dimensional Euclidean survey geometry: when a telescope can only detect celestial objects brighter than a limiting apparent magnitude m_lim (minimum flux F_min), a star or galaxy on the brighter side of the population's mean absolute magnitude M₀ remains visible out to a larger maximum distance d_max(M) = 10^{1 + 0.2(m_lim − M)} parsecs—and therefore across a survey volume V_max ∝ d_max³ ∝ 10^−0.6M that is much larger than for an otherwise identical object on the fainter side of the mean. Consequently, the average absolute magnitude ⟨M⟩_obs of the objects captured in a flux-limited catalog is brighter (more negative) than the volume-limited population mean M₀ by an amount proportional to the variance σ_M²:

Δ⟨M⟩ = ⟨M⟩_obs − M₀ = − (ln 10) × 0.6 σ_M² ≈ −1.382 σ_M² (Uniform Spatial Density)

If an astronomer assigns the volume-limited mean luminosity M₀ to objects drawn from a flux-limited sample without correcting for this selection bias, all inferred distances are underestimated by a fractional factor exp(−0.691 [ln 10]² × 0.04 σ_M²)—an error that historically inflated early twentieth-century measurements of the Hubble constant (H₀) and continues to require Bayesian modeling in Type Ia supernova cosmology, Tully–Fisher galaxy peculiar-velocity surveys, and gravitational-wave standard-siren analyses (Malmquist, 1920, 1922; Lynden-Bell et al., 1988; Strauss & Willick, 1995; Teerikorpi, 1997).

Consider a class of astronomical objects (such as main-sequence stars of a given spectral type, Cepheid variables at a fixed period, or spiral galaxies at a fixed rotational velocity width) whose true volume-limited luminosity function φ(M) follows a Gaussian distribution with mean absolute magnitude M₀ and intrinsic dispersion σ_M:

φ(M) = (1 / √(2π σ_M²)) exp[ − (M − M₀)² / (2 σ_M²) ]

At distance r (in parsecs, neglecting extinction), an object of intrinsic magnitude M has apparent magnitude m = M + 5 log₁₀(r / 10 pc), or r(m, M) = 10^{1 + 0.2(m − M)}. If the spatial number density of objects at distance r is n(r), the joint number of objects observed inside solid angle ω within apparent magnitude bin [m, m + dm] and magnitude bin [M, M + dM] is:

d²N(m, M) = ω r² n(r) (∂r / ∂m) φ(M) dm dM = (0.2 ln 10) ω r³ n(r) φ(M) dm dM

Integrating over all magnitudes M yields the differential star-count function A(m) = dN(m)/dm (the number of objects per unit apparent magnitude at apparent magnitude m). Using Bayes' theorem and completing the square in the Gaussian exponent, Malmquist (1920, 1922) derived his general relation for the conditional mean ⟨M | m⟩ of stars observed at a given apparent magnitude m for any spatial density profile n(r):

⟨M | m⟩ = M₀ − σ_M² ( d ln A(m) / dm )

In the classical homogeneous case where objects are distributed with constant spatial density n(r) = n₀ throughout Euclidean space, the cumulative star count up to apparent magnitude m scales with volume as N(< m) ∝ r_max³ ∝ (10^0.2m)³ = 10^0.6m, so that A(m) = dN/dm ∝ 10^0.6m = exp(0.6 [ln 10] m). Therefore, the logarithmic slope of the star counts is a universal constant:

d ln A(m) / dm = 0.6 ln 10 ≈ 1.38155

Substituting this slope into Malmquist's general formula yields the Classical (Homogeneous) Malmquist Bias:

ΔM_Malmquist = ⟨M⟩_obs − M₀ = −0.6 (ln 10) σ_M² ≈ −1.382 σ_M² magnitudes

Furthermore, Malmquist proved that the conditional distribution of absolute magnitudes at fixed m (or for m ≤ m_lim) remains a Gaussian with the same variance σ_M², shifted brighter by −1.382 σ_M².

• Quantitative Impact on Distance Estimates: Suppose a class of standard candles has M₀ = −19.0 mag and an intrinsic scatter of σ_M = 0.40 mag. In a magnitude-limited catalog, the average observed object actually has ⟨M⟩_obs = −19.0 − 1.382(0.40)² = −19.221 mag (~22% more luminous than M₀). If an observer assumes the object has M₀ = −19.0 mag, its inferred distance modulus μ = m − M₀ is underestimated by 0.221 mag, underestimating its true distance by ~10%.

In modern astrostatistics and cosmology, three closely related biases are distinguished (Lynden-Bell et al., 1988; Teerikorpi, 1997; Luri et al., 2018):

During the 1970s and 1980s, Malmquist bias stood at the center of the famous "factor-of-two controversy" over the Hubble constant (H₀):

• Allan Sandage and Gustav Tammann argued for a "long distance scale" (H₀ ≈ 50–55 km s⁻¹ Mpc⁻¹), demonstrating via Spaenhauer diagrams (plots of absolute magnitude M versus log distance log r) that galaxy catalogs compiled by Gérard de Vaucouleurs (H₀ ≈ 90–100 km s⁻¹ Mpc⁻¹) suffered from unmodeled Malmquist bias: as distance r increased toward the catalog flux limit, only galaxies from the upper luminosity envelope of each morphological class remained detectable, compressing the apparent distance scale by nearly a factor of two (Sandage, 1994; Teerikorpi, 1997).
• Conversely, using inverse Tully–Fisher regressions—fitting line width η = log(2v_max) as the dependent variable against magnitude M (since galaxy selection is normally cut on magnitude rather than HI line width, Schechter, 1980)—together with space-based Hubble Space Telescope Cepheid observations that pushed the flux limit m_lim deep below the lower tail of the Cepheid period–luminosity relation, the HST Key Project (Freedman et al., 2001) and SH0ES Collaboration (Riess et al., 2022) eliminated classical Malmquist bias on the local rung of the cosmic distance ladder.
• Today, in gravitational-wave cosmology, a direct analog of Malmquist bias—selection bias toward face-on orbital inclinations (i → 0° or 180°), because face-on binary neutron stars and black holes emit gravitational waves with √(1 + 6 cos²i + cos⁴i)/2 = 2.83× stronger strain amplitude and are thus detectable across 22.6× larger cosmic volume than edge-on binaries—must be marginalized when measuring H₀ from LIGO–Virgo–KAGRA standard sirens (Schutz, 1986; Abbott et al., 2017).

1. Eddington, A. S. (1913). On a formula for correcting statistics for the effects of a known probable error of observation. Monthly Notices of the Royal Astronomical Society, 73(5), 359–360. https://doi.org/10.1093/mnras/73.5.359
2. Malmquist, K. G. (1920). A study of the stars of spectral type A. Meddelanden från Lunds Astronomiska Observatorium, Serie II, 22, 1–69. https://ui.adsabs.harvard.edu/abs/1920MedLu..22....1M
3. Malmquist, K. G. (1922). On some relations in stellar statistics. Arkiv för Matematik, Astronomi och Fysik, 16(23), 1–52. https://ui.adsabs.harvard.edu/abs/1922ArMAF..16...23M
4. Lutz, T. E., & Kelker, D. H. (1973). On the use of trigonometric parallaxes for the calibration of luminosity systems: Theory. Publications of the Astronomical Society of the Pacific, 85(507), 573–578. https://doi.org/10.1086/129506
5. Lynden-Bell, D., Faber, S. M., Burstein, D., Davies, R. L., Dressler, A., Terlevich, R. J., & Wegner, G. (1988). Spectroscopy and photometry of elliptical galaxies. V. Galaxy streaming toward the new supergalactic center. The Astrophysical Journal, 326, 19–49. https://doi.org/10.1086/166066
6. Strauss, M. A., & Willick, J. A. (1995). The density and peculiar velocity fields of nearby galaxies. Physics Reports, 261(5–6), 271–431. https://doi.org/10.1016/0370-1573(95)00013-700013-7)
7. Teerikorpi, P. (1997). Observational selection bias affecting the determination of the extragalactic distance scale. Annual Review of Astronomy and Astrophysics, 35, 101–136. https://doi.org/10.1146/annurev.astro.35.1.101