I'm writing a high school paper on the bioremediation of ZnSO4 solution using yeast (S. Cerevisiae), where the independent variable is temperature (15, 25, 35, 45 and 55 degrees celsius). One of my teachers said that it doesn't make sense to find overlapping error bars and use t-tests to measure statistical significance because, as my data is continuous, I am supposed to look for an overall trend rather than differences in the resulting dependent variable between each variation of the IV. Another teacher said I should use a t-test for the data points with overlapping error bars to confirm whether the difference is statistically significant. Which is the correct advice?
$\begingroup$
$\endgroup$
3
-
$\begingroup$ The question cannot be answered without more details about the experiment and the research question. $\endgroup$– ocramCommented Feb 12, 2021 at 15:03
-
$\begingroup$ @alice Please register &/or merge your accounts (you can find information on how to do this in the My Account section of our help center), then you will be able to edit & comment on your own question. $\endgroup$– Sycorax ♦Commented Feb 12, 2021 at 16:04
-
$\begingroup$ It's helpful to think about the purpose of the experiment. In biology, we typically use a temperature gradient to identify or approximate the optimum temperature for a given metabolic process (in your case, zinc sulfate remediation). Using t-tests to compare each discrete group tells you nothing about what might be happening in between those temperatures. It's far more interesting and useful if we use a regression analysis to identify the trend between temperature and your dependent variable. $\endgroup$– MikeyCCommented Feb 16, 2021 at 15:53
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
Since you're in high school, I'm going to break some of my own rules and give you some broad advice.
The t test is really appropriate in two settings:
- You have two groups and the outcome is continuous.
- You have more than two groups and prior to collecting data, you decide which set of groups you want to test against each other.
From the details you give us (I'm assuming bioremediation of ZnSO4 solution using yeast is a continuous outcome), you should not do a t test. Why?
Your independent variable (temperature) is a continuous variable in reality, not a discrete variable (i.e. it can not be considered a "grouping" variable). It would be reasonable to ask about bioremediation at a temperature between 15 and 25, and you would likely want to use the data you have to estimate an outcome for that temperature.
Instead, you should use something like linear regression. If your goal is to find a trend, then linear regression naturally can test for that.
Additionally, overlapping confidence intervals do not mean that there is no statistically significant difference. If the confidence intervals fail to overlap, then this would indicate significance, but confidence intervals can overlap and their difference can still be significant.
answered Feb 12, 2021 at 16:03