English translation: worst case scenario analysis or approach
Login or register (free and only takes a few minutes) to participate in this question.
You will also have access to many other tools and opportunities designed for those who have language-related jobs (or are passionate about them). Participation is free and the site has a strict confidentiality policy.
French to English translations [PRO] Science - Mathematics & Statistics
French term or phrase:Hypothèse du biais maximum
This occurs in an observational study of patients receiving artificial heart valves. The specific section deals with how the data is to be analyzed. Here it is:
Analyse des données disponibles avec, pour le critère principal, la possibilité d’effectuer une analyse de sensibilité sous ***l’hypothèse du biais maximum***, en considérant comme morts les patients perdus de vue. Analyse basée sur le principe « d’intention de traiter » sur l’ensemble des patients inclus.
I don't really get far with the logical 'maximum bias (or maximally biased) hypothesis'. The only lead I have is a definition, which is again in French :
Hypothèse du biais maximum:
Hypothèse dans laquelle on choisit de se situer, au moment de l'analyse, dans la situation la plus défavorable,
pour arriver à conclure à une différence.
So I basically follow the concept that lost to follow-up patients are considered 'dead' as a worst-case scenario.
Is there any statistical wizard out there that can enlighten me?
Many thanks in advance!
Explanation: Speaking as an epidemologist, I really don't think you need to (or should) include the word "bias". It's a different way of looking at things.
after reading your reference, that you've found a context (clinical studies) where the researchers are so acutely aware of bias, and so concerned with eliminating it, that the scenarios in the sensitivity analysis are all about bias, and the "worst case scenario" is the worst case for the validity of the *clinical study* -- not for patient outcomes.
Which is not to say that the French author's less judgmental, factually descriptive way of characterising the assumption (as biased as can be) ought to be left out altogether in the translation. If the word 'biais' appears in the source text but "bias" is found nowhere in the translation, the author may not feel well served. Bias is a key consideration.
Also, if you don't use the word "bias" somewhere, you aren't telling the reader *in what way* the assumption assumes a worst case. The French text does this. The issue is making a sensitivity analysis that is meaningful even when the estimator is biased.
ML and bias are two different characteristics of an estimator. MLEs can be subject to bias and corrected for it.
Yes, this is a worst-case scenario, but for the *estimation technique*: the assumption is that the bias of the estimator is as bad as it could be. The problem is that usually in a sensitivity analysis, the worst-case scenario is the one likely to produce the worst outcome (here, I assume, for the patients). And it that were intended, you'd expect to see something like hypothèse du scénario le plus pessimiste/scénario catastrophe.
@rkillings: I won't use "maximum bias" since I can't find any evidence of its use in a case like this. The investigators want to find a way of minimizing the negative effects of missing data (i.e., lost to follow-up patients), which could skew the results. There are plenty of examples and references to this method for dealing with missing data on the internet from reliable sources and I am convinced it is the 'worst case scenario'.
@ Rachel: I came across the same term but don't really see how that would fit in here. It is a method for determining a "likely" outcome based on a few samples of the population. WCS analysis deals with the missing data with the aim of reducing the bias it would cause.
That's what it says in French, and as it stands, it's a statement about the *estimator* -- assuming the worst case for the accuracy thereof -- not about the condition of the patients.
I really appreciate the educated input of both you and Sue. It's been years since I studied stats as well and frankly, I'm just too tired to delve into it too deeply, at least tonight ;)
I like the idea of a compromise however....all bases covered then I guess.
IMO, the author(s) probably consider the way in which they treat bias to be important, so the word itself should not be discarded. Maybe "a sensitivity analysis in which a "worst case" bias is assumed"? Splitting hairs, perhaps?
The book referred to makes some interesting observations on "publication bias".
Explanation: It's a long time since I did any "proper" statistics, but bias is all-pervasive in real experiments. Note that "hypothèse" is frequently more accurately translated as "assumption", so here the experimenter is (presumably) assuming a worst-case scenario which includes the greates possible bias in his data, leading to a rather conservative outcome or prediction.
See http://en.wikipedia.org/wiki/Selection_bias for a useful discussion of bias.
chris collister France Local time: 20:45 Specializes in field Native speaker of: English PRO pts in category: 67
Explanation: Speaking as an epidemologist, I really don't think you need to (or should) include the word "bias". It's a different way of looking at things.
SJLD Local time: 20:45 Specializes in field Native speaker of: English PRO pts in category: 48
Grading comment
Thank you so much Sue for putting me on the right track :) And thank you to the others for their input.
Explanation: The "worst case bias" part is common: ~13,500 Ghits. Not all of them relate to the (several, different) meanings of "bias" in statistics, but on inspection, many do.
In estimation, less bias is always good, so it's easy to equate "worst" and "maximum".
rkillings United States Local time: 11:45 Works in field Native speaker of: English PRO pts in category: 24
The translation workplace 
