【英語論文の書き方】第92回　「誤解を招く指標を避けるために、何を、なぜ測っているのかを理解する」について 論文翻訳・英文校正　研究者専門の翻訳会社　ワールド翻訳サービス

第91回では「*誤った*精度：パート2：統計的確率」を取り上げました。
 
第92回（今回）のテーマは
「誤解を招く指標を避けるために、何を、なぜ測っているのかを理解する」
についてです。
 
科学的思考の美点の一つは、主観的な評価を、
より客観的で比較しやすい定量的な測定値で置き換えようとする点にあります。
 
このアプローチは、医学のような人間中心の分野から
化学のような、より機械論的な分野まで、
多くの領域で大きな進歩をもたらしてきました。
 
その一方で、このアプローチは、
重要な事柄を見落とさせてしまうこともあります。
最も重要なことの一つは、
「測定可能である（あるいは数値化できる）というだけで、
その数値に意味がある」と考えてしまう、誤った信念です。
 
Geoffさんは、その例として、
広く用いられている医学指数であるボディマス指数（BMI）や、
医療で用いられる疼痛スケールを取り上げ、
本来の目的や文脈から外れて使われた結果、
「garbage metric(意味をなさない指標)」になってしまう
危険性を指摘しています。
 
また、サイエンスエディターとして仕事をするなかで、
著者が定義した尺度の方向性が変数名の意味と逆行している事例や、
そのために自身のデータを誤って解釈してしまう事例を
頻繁に目にすると言及しています。
 
数値の意味は常に文脈に依存しており、
指標を作る際にその文脈を忘れてしまうのは賢明ではない、
というのがGeoffさんの意見です。
論文を書く際は、著者自身と読者にとっての
数値の意味を十分に理解したうえで、
自分が何を伝えようとしているのかを
慎重に検討することが大切だとも述べています。
 
本記事が、数値や指標との向き合い方を考える
きっかけとなれば幸いです。

 
    One virtue of scientific thinking is that it tries to replace subjective assessments with quantifiable measurements that are more objective and easier to compare. This approach has greatly advanced many areas of endeavor, including both human-centered fields such as medicine and more mechanistic fields such as chemistry. This approach has been tremendously beneficial, but can blind us to important things. One of the most important is the incorrect belief that just because you can measure something (or assign a number to it), the number is meaningful.
 
Note: Measurements, or indexes (also called indicators) based on measurements, are called metrics, from the Greek word metreo (“to measure or compare”).
 
    Consider the example of a widely used medical index, the body mass index (BMI). This index was designed to determine whether a population of individuals has a generally healthy body weight or is (on average) obese, so that its members need to lose weight. The index is calculated as follows:
    BMI = weight (kg) / height (m) squared
    You can calculate your own BMI using the online calculator provided by the U.S. National Institutes of Health: (https://www.nhlbi.nih.gov/health/educational/lose_wt/BMI/bmicalc.htm). But why would you? BMI was not designed to describe an individual’s status; rather, it is a population-level measure and is therefore of questionable value for individuals. Examining the two parts of the indicator reveal the problem. First, relying only on weight, without qualifying the weight’s nature, ignores the factors that contribute to weight. It therefore fails to distinguish a robustly healthy individual, such as my female friend who is a champion powerlifter, and someone who’s clinically obese. That is, the equation treats 1 kg of muscle as identical to 1 kg of adipose tissue (“fat”). Second, height is a linear measure, which is then squared to represent an area measure. Yet bodies are three-dimensional, and height is a poor proxy for that complexity. Squaring the height term exacerbates this problem.
    The result is what I call “a garbage metric”—one that is applied outside the purpose for which it was designed or poorly designed in the first place. It seems to measure something important, but only indirectly describes the thing being measured. BMI can indeed reveal something important, such as the consequences of poor diet and insufficient activity in a population. But is not, by itself, sufficient to confirm the problem in an individual.
 
Note: I chose the term “garbage” based on the famous computer science maxim that if your input data is “garbage”, your output (conclusion) is also likely to be garbage.
 
    Consider a second example, which most of us have experienced during a visit to our doctor to mitigate pain. Doctors generally ask us to describe the pain’s intensity. In North America, this is commonly done using a pain scale from 1 (a minor annoyance) to 10 (the worst pain the patient can imagine). The problem should be clear: if you’ve been fortunate in living a largely pain-free life, the amount of pain you can imagine is limited, whereas if you suffer from chronic severe pain, you can imagine pain that a pain-free individual could never imagine. If you’ve experienced a kidney stone or childbirth, you can imagine far worse pain than someone whose worst pain was caused by a bruise or headache. Moreover, the most common version of this scale lacks definitions for intermediate values, making it impossible to reliably assign intermediate values to a patient; asking the same question about exactly the same amount of pain will result in different numerical values each time. Another garbage metric!
    Consider, instead, a pain metric that is based on the impact on the patient. This revised scale could use the same range of values (0 to 10), but might include the following categories:
· 0 = no pain
· 1 to 3 = I only noticed the pain because you asked.
· 4 to 7 = I feel enough pain that I cannot perform a normal activity, such as bending over to touch my toes.
· 7 to 9 = The pain is so severe that I am having a hard time breathing.
· 10 = Pain so intense that the patient couldn’t respond to the question.
    This metric is much more useful because, in addition to clearly defining the meaning of the categories (so that asking the same question repeatedly will obtain approximately the same answer each time), it expresses the pain in terms that are consistent across patients and meaningful because they reveal the pain’s impact. After all, the goal of the question is to learn what the pain means to the patient so you can decide how aggressively to treat it.
    Another problem relates to the implied meaning of a metric. When you choose variable names, ensure that they reflect the meaning of the data. For example, many researchers use the conventions of binary logic to divide results into two categories, with 1 = yes or true or present and 0 = no or false or absent. This works well for binary choices (e.g., treated for pain versus not treated) because 0 represents the absence of a value, so any non-zero value must mean the presence of a value. (This would not be true for values that have multiple values between 0 and 1.0, such as correlation coefficients.)
    There are potential problems with variable names that contradict the meaning of the variable’s value. If you use a variable named “response strength” with values ranging from 1 to 10, it’s clearer to use 1 (a low number) for a low strength and 10 (a high number) for a high strength. Using 1 = the highest strength and 10 = the lowest strength, an order used in athletic competitions is contradictory because a low score for an athlete (1) is associated with a high value (the highest strength or skill). This kind of mismatch between the variable name and the variable’s value encourages misunderstandings and interpretation errors. In my work as a scientific editor, I often see authors interpret their own data incorrectly because the scale they defined runs in the opposite direction to the meaning of the variable name.
    The meaning of a number always depends on its context, and it’s unwise to forget that context when you develop metrics. Think very carefully about what you’re hoping to communicate. The fact that you’re creating a number is, by itself, unimportant. Without understanding the number’s meaning to yourself and your readers, you’re at high risk of creating a garbage metric.

メールアドレス
お名前

メルマガ登録