hi,
i'm writing a research about decolorization of waste water by fungi. the results obtained by % decolorization. Is it necessary to include test of significance??
Suppose you have 16% decolorization and reported 16%, is this report meaningful? No, it is not meaningful at all because it has no context. By reporting the pValue, it puts the result of 16% in context; that context is (i) the standard decolorization (in this case) and the standard deviation of the past observation here the current 16% is compared.
A numerical illustration would be helpful Suppose we have the standard figure obtained through past observation as the reference to which the current 10% is compared. The past observations are:
15%
11%
12%
13%
14%
The mean is 0.13 and SD = 0.02. If your current observation is 16%. Without report the pValue, this 16% is not that meaningful. To be more meaning, we ask, what is the significance level for 16%?
Z = (Xobs - Xmean) / SD
Z = (0.16 - 0.13) / 0.02
Z = 1.90
Now go to the Z table and find the corresponding % probability for critical value of 1.90. The answer is F(X) = 0.971 from the Z table. pValue = 1 - F(X) or 1 - 0.971 = 0.029. Compare to 5%, this is a significant finding pValue = 0.029.
If you simply want to describe what you found in your experiments, there is no need for significance tests.
The problem starts when you want to claim what one would expect to find, if one would repeat your experiments. That means, you can find that *your* funghi in *your* experiment showed some decolorization, but you would claim that this is repeatable by others using *similar* funghi in *similar* conditions. This is usually why we do research... Noteably, the results in a replicated experiment won't be eaxctly the same as in *your* experiment. Thus, *your* 16% may possibly a very bad estimate of what others might find. You should therefore somehow indicate how variable you would expect the results of others to be, given your experince (i.e. considereing the variability of *your* data). There are many simple and approximate methods. For instance, you could simply report the range, an interpercentile range, the variance or the standard deviation of *your* values. That gives a rough impression of what to expect in replications. Giving a some more precise expression of the "repeatability" is rather difficult and actually nobody does it. Instead, people fall back to significance tests. These tests are only about the sign of the effect. *Your* effect is positive (in terms of "decolorization"), so it is about that the colorization decreases rather that it increases. The test makes no claim about the size of the effect. That is, it is irrelevant if *your* estimate is 16% or 4% or 99% - it's only about the sign. If the test is "significant" you claim that your data is sufficient to interpret the sign of *your* estimate. If it is not significant, you can't even interpret the sign. It's no statement about repeatability and about what others are expected to find. It's only a statement about *your* data, that the "signal-to-noise" ratio in *your* data is large enough to at least interpret the sign of *your* estimate. This is what you would need the p-value for, and this is a bare minimum requirement if you want to talk about *your* data (whether there is a positive or a negative effect -- or if you can't see that).
Better than the p-value is the confidence interval. It is also only about *your* data (just like the p-value), but it gives a range of estimates that are "not too incompatible" with *your* data.
I think if your aim is to just describe the decolonization level, there is no need for the test of significance (p value). However, if you want make inferences out of the results then it is necessarily and in this case I suggest you go with Paul Louangrath's opinion.
- Badges
- Science topic
- Similar topics
- Linguistics
- Composition Studies
- Writing