eLetters to:

ARTICLES: Robert E. Lasky and Amber L. Williams Noise and Light Exposures for Extremely Low Birth Weight Newborns During Their Stay in the Neonatal Intensive Care Unit Pediatrics 2009; 123: 540-546 [Abstract] [Full text] [PDF]

eLetters: Submit a response to this article

eLetters published:

Logarithm

Renaud Vialet   (31 July 2009)

Logarithm 31 July 2009
Renaud Vialet, Physician Réanimation Pédiatrique, Assistance Publique, Hôpitaux de Marseille, Hôpital Nord Send letter to journal: Re: Logarithm renaud.vialet{at}ap-hm.fr Renaud Vialet	To the Editor: We read with great interest the recent paper by R.E. Lasky and A.L. Williams1 studying sound and light exposure, which are very important factors affecting newborns in neonatal intensive care units. We only did not completely agree with the mathematical transformations the authors made before statistical analysis. The authors stated that "Statistics calculated on sound measurements recorded in decibels (dB) without transforming them are misleading. [… and …] the dB measurements must be converted to sound pressure, a linear scale, before averaging". Actually, dB is the commonly used unit for sound measurements, and is a (Neperian) logarithmic transformation of Pascals (Pa). In facts, a logarithmic transformation is useful to make calculations easier, as multiplications are simplified in additions. So logarithmic transformations should be used with precautions: for example if you add up sounds; for example, a friend calling out to you (70 dB) in a noisy restaurant (70 dB either), for you the sound will be 70 dB + 70 dB, which equals (approximately)… 73 dB, and not 140! . As stated by the authors, for that calculation, you must before addition transform dB in Pa, and after re-transform Pa in dB. However, these transformations are not necessary for statistical analysis. Commonly used statistical analysis, like those used in the paper (calculations of confidence interval, generalized linear mixed models analysis) make conditions not on the variable itself but on the distribution of the variable, which must be normal (Gaussian). Let us use a comparison with a variable more familiar to the clinicians, the pH. pH is a (decimal) logarithmic transformation of the concentration of H+. For the same reasons than above, if you blend the same volume of a pH=6 solution with a pH=8 solution, you will not obtain a pH = 7 solution, but a (approximately) pH = 6.3 solution. However, as blood pH of our patients follow a normal distribution, it is correct when performing statistical analysis of the results, to use statistic tests without any prior mathematical transformation, as it is actually done in all the medical scientific articles. And it is quite the same for dB analysis. Reference: 1: Their Stay in the Neonatal Intensive Care Unit, Noise and Light Exposures for Extremely Low Birth Weight Newborns: Pediatrics 2009;123;540 -546 Conflict of Interest: None declared