I basically want to see if there is a difference between these sites. Especially between the forehead and inner wrist as many patients are not very happy with temperature reading taken at the forehead with non contact infrared thermometers (NCIT). Should I use Bland-Altman plot, t -test or a correlation test ? I have readings from single NCIT totaling 1000 (my sample size). However my calculated sample size is 325.
Dear Tahirusevli,
If the distribution of your data is normal, you could use one way ANOVA as a statistical test.
Sorry, AmirAli Jafarnezhadgero ANOVA is no use here, at least not standard ANOVA, because this only compares mean values.
What you need is to compare the individual readings. Two methods of measurement could have the same mean, but very different values for each pair of measurements.
There are two ways of conceptualising agreement. The first is absolute agreement : this means that the two methods should have the exact same reading. The second is consistency : this means that the two methods of measurement rank the observations in the same order. This is useful for arbitrary scales such as ratings of job candidates, but not really for values like temperature.
You can plot the disagreements between two methods of measurement using Bland and Altman's method (see https://pubmed.ncbi.nlm.nih.gov/2868172/). This tells you the average disagreement and the range into which 95% of disagreements are expected to fall.
You can measure the agreement more formally with an intraclass correlation. With temperature measuring devices, the devices are considered fixed factors (with two or more raters, the raters would be considered random factors). In this case, you need a type II intraclass correlation. And you want an intraclass correlation for absolute agreement.
Ronán Michael Conroy gave the best answer
I have three more suggestions. Maybe they are wrong!
1. Repeated measure ANOVA.
2. Pearson correlation.
3. GEE as univariable in which the only independent variable is the repeated measure variable (as factor var with oral temperature as base). The resulted beta coefficient can be used to create a correction factor (to estimate oral temperature by proxy). In addition, age and sex can be added to the model to design a correction formula.
Hello Thiruselvi,
If you simply wish to know whether site (forehead vs. wrist) for a NCIT measure yields different results, then I concur with Safaa K. Kadhem and Seyyed Amir Yasin Ahmadi 's comments that either a paired/dependent/correlated t-test or simple repeated measures ANOVA would suffice to answer your question. A confidence interval for the difference, as hinted at by Ronán Michael Conroy 's response, should be easy to report from either analysis, and would be helpful for your audience.
Good luck with your work.
Dear David Morse and Ronán Michael Conroy , sorry I disagree with you both. I think independent T-test should be used in this case.
Hello all,
Here's why I think the independent t-test is not the best choice (as suggested by Robel Hussen Kabthymer ). There is very likely a correlation between temperatures taken at different sites from the same persons. The dependent-t and repeated measures anova take advantage of this correlation to test the null hypothesis of no difference across (within-subject) conditions (in this instance, sites for temperature measurements) with greater sensitivity than the independent-t (or ordinary one-way anova).
How does this happen? The independent t uses person differences to estimate the "error" term for comparing observed mean differences, whereas the dependent t uses the variation of score differences (which is guaranteed to be smaller than than the independent t error as long as there is a positive correlation between scores). The smaller error term typically yields more sensitive (a.k.a., powerful) tests, despite the difference in error df.
If you like, try analyzing the attached data set both ways and decide for yourself.
Best regards
The forehead and wrist temperatures taken from the same person will be different. Even the right and left tympanic temperatures will be different. So you certainly would not want to assume that the mean difference in temperatures across individuals is 0.
Instead you might consider whether the temperatures are correlated. I would start by examining a scatter plot of forehead and wrist temperatures. If the two temperatures appear to be linearly related, you test the correlation between them. On the other hand, the scatter plot might show an interesting non-linear relationship between temperatures taken from the two different sites.
This is the first time I am using this platform and I am glad I did. Thank you everyone for the feedback and references. I did not expect this many differences in opinion.
The ultimate objective of this research is to be able to take the temperature using NCIT at the wrist if the patient/person is not keen for the forehead. In this study we used the same NCIT for all the participants at the all the 3 sites. I did consider an indpd t test initially but was also advised about Bland-Altman.
- Badges
- Science topic
- Similar topics
- Psychology
- Emotional Psychology
- Happiness