John Peterson Oct 27, 2008:

Last comment on this I don't think that this is the place to go into a semantic and theoretical discussion of sampling theory, so this is the last point I'll make. In a nutshell, in tests like this there are two hypotheses that you can put forward for the outcome/result obtained: either the difference is due to chance (null hypothesis) or there is a (statistically) significant difference between men and women (within the population as a whole). The decision is framed in terms of rejecting the null hypothesis. Thus there are thus two probabilities: the probability of taking the wrong decision (p) and the probability that the decision that there is a statistically significant difference is the right one (1-p). Taking a 95% confidence interval you are saying, in effect, that you are 95% sure that the decision to reject the null hypothesis in favour of a decision which says that the difference is statistically significant is correct. In other words, in rkillings' approach, the selected confidence interval means that the probability of this decision being a Type 1 error (rejecting the null hypothesis when it is, in fact, true) is 5% (p=0.05) - 5% being the area of the normal curve lying outside the 95% confidence interval boundary. Crudely, one (95%) implies the other (p=0.05) - as the sum of the two probabilities, by definition, equals 1. If the author has put it in terms of 95% confident that the outcome is statistically significant(and is broadly right - (s)he may not be an expert statistician) then my suggestion would be stick with that approach and use the wording suggested by BDF or myself.

John Peterson Oct 27, 2008:

To Ormiston - I think that you can keep things this way, if that how the author has put it. See note in discussion box.

B D Finch Oct 27, 2008:

Re rkillings' disagree with John Peterson Sorry, but rk is mistaken. He is confusing significance with confidence (easily done). A glance at the link in the Wikipedia article he cites will show the problems with p-levels. Also see http://en.wikipedia.org/wiki/Confidence_level for the 95% (or any other) level of confidence/confidence level.

ormiston (asker) Oct 27, 2008:

to rkillings I bow to anyone's superior knowledge here - how do you therefore think this should have been phrased ?

John Peterson Oct 25, 2008:

Keep it simple Just to second BDF's comments about not including a reference to the sd and to use something like her suggestion or mine. It is easy to make a slip sometines (like the 1.96) and so run the risk of that slip being carried into the translation, so something that just mentions "95% confidence level" will do fine.

John Peterson Oct 25, 2008:

You're right Andy - thanks you're right it should be +/-1.96sd for 95% of the area - shows what happens when you type from from memory in a hurry!

B D Finch Oct 25, 2008:

Do not mention SD It is useful for the translator to understand the theoretical underpinning of levels of confidence. That is where the explanation of SDs comes in. This should not appear in the translation. Oh, and the 95% level is not set in stone. If we were talking aeroplane safety, I'd prefer a 99% level before agreeing to fly. Sociological studies sometimes opt for a 90% level.

B D Finch Oct 25, 2008:

+/-1.96? In response to Andy's question to John: no, it should not read "+/-1.96"! That would be both confusing and wrong. The French doesn't give a stats lesson, so the translation shouldn't either. It just needs to be correct and "statistically significant at a/the 95% level of confidence/confidence level" is all that's needed.

Gayle Wallimann Oct 24, 2008:

Discussion moved from "clarification box" John Peterson: 11:23am Oct 24, 2008: I think I'd stick at "the 95% confidence level if you can". However, terms like "95% certain" are used as well - so that might help. As for being Greek, the symbol for the standard deviation (crudely, a measure of "spread") is the lower case sigma! [Hide]
ormiston: 11:27am Oct 24, 2008: am gratefully trying to assimilate all this - but how do I deal with this embellishment?!
"écart significatif à 95% entre les hommes et les femmes" [Hide]
John Peterson: 11:42am Oct 24, 2008: Maybe say something like: "we can be 95% certain that the difference between men and women is (statistically)significant". - I'd try and keep the "statistically". Or: "There's a 95% certainty that ..." [Hide]
B D Finch: 1:42pm Oct 24, 2008: I would disagree with JP's last posting. If one adopts a 95% level of confidence (and the level should be selected before the results are known) then one is saying that the result above that **IS statistically significant** , it might still be wrong. [Hide]
B D Finch: 1:47pm Oct 24, 2008: In other words, one can know that something is statistically significant. At a 95% probablity level, that makes it 95% likely to be true, however, it is still 5% probable that it is untrue. Even if untrue, the statistical difference is still significant [Hide]
John Peterson: 1:54pm Oct 24, 2008: Re BDF's comment - as we're talking about probabilities, the decision to reject the null hypothesis can be wrong; but the assumption is that a high confidence level reduces the chance of making the wrong decision in terms of rejecting the null hypothesis. [Hide]
John Peterson: 2:05pm Oct 24, 2008: The above was a response to BDF's 1st point. Re the second point, statistical significance is about the chance of being right/wrong. The difference may be numerically significant (big/small) but not statistically significant (likely to be right). [Hide]
ormiston: 3:43pm Oct 24, 2008: thank you for this help! - my head is spinning. I have no room for a lengthy sentence - can it be made any neater than this ? (!)
"statistically significant difference at the 95% confidence level from the +/- standard deviation" [Hide]
John Peterson: 4:03pm Oct 24, 2008: I'd just say "the difference is statistically significant at the 95% confidence level". In my view, you don't need to mention the standard deviation - it's implied by 95% being equal to an area bounded by +/- 1 sd of a standardised normal distribution. [Hide]
Andy THEODOROU: 8:54pm Oct 24, 2008: Hello John - should that read +/-1.96... ? [Hide]