Research methods encompass an immense amount of knowledge that is related, but tangential, to the core knowledge necessary to be a practicing pediatric hospitalist. Mastering these enormous fields as a practicing physician is almost impossible, which is why physician-scientists work closely with statisticians and other methods experts. However, it can also be overwhelming to those who want to learn more about research methods. Further, some may lack access to people with the relevant content expertise who can help.

Method/ology is here to help. As a break from our typical fare, we will occasionally feature a piece called “Dear Methodologist” where readers of Hospital Pediatrics can submit an anonymous question for consideration that will be answered by experts. This is the first in this series, meant to be a light-hearted but informative (and brief) read, which will appear sporadically as questions are received.

Dear Methodologist:

I recently attended the annual Pediatric Hospital Medicine (PHM) conference, where I saw a lot of great content. However, I noticed terms that were used in several different ways, including some that are part of common language but seem to mean something different in a research context. I wasn’t entirely sure that I actually understood the meaning of these terms, which differs from what I can find in a standard dictionary. Could you help me understand? The words in question are: significant, correlation, trend.


Evidence-Based Medicine Confusion

Dear Evidence-Based Medicine Confusion –

Thank you for this question! I am happy to help, as there is much confusion around the proper use of these terms. As is true in any field, precise use of language is critical to both conveying and understanding information correctly and terms that also hold related, but subtly different meaning when used in everyday language, causes additional confusion.

Correlation: Correlation (or corelation, if it helps) is a term used to discuss 2 continuous variables that are linearly related. We will be referring to linear correlations here. These relationships are either positive (ie, as 1 variable increases, so does the other) or negative (Fig 1, panels A and B). An example of a positive correlation is the number of patients on a PHM service and the time that it takes to round on service: the more patients there are, the longer it will take. “Correlation” refers to a more specific type of relationship than association, which is a more general term. A similar example of an association that is not a correlation is medical complexity of the patients on the PHM service and the length of rounds. “Medical complexity” is typically a categorical variable and not a continuous variable (yes, I realize that there are likely ways to assess this continuously, but let’s not be too pedantic here), and thus you can’t use a measure of linear correlation to look at the relationship. Correlation is measured statistically using different types of tests, the most common of which is Pearson (which can be indicated by an “r” in relevant articles). The results of these tests will tell you the strength of the linear relationship, where 1 (or −1, for a negative correlation) is a perfect relationship and 0 is no relationship. Sometimes 2 continuous variables have a relationship that is curvilinear (ie, the ratio of change is not constant) and not linear (Fig 1, panel C). In these cases, the usual measures of linear correlation, like Pearson correlation, may not give you accurate information. Given the specificity of this word in scientific writing, it is preferable to use a word such as “association” to avoid confusion, even if the word correlation itself might have made sense in another context.


Examples of positive linear, negative linear, and a curvilinear correlation. A, Positive linear correlation. B, Negative linear correlation. C, Curvilinear correlation.


Examples of positive linear, negative linear, and a curvilinear correlation. A, Positive linear correlation. B, Negative linear correlation. C, Curvilinear correlation.

Close modal

Trend: This is a term that is commonly heard during scientific presentations, in which a presenter may say, “the results trended towards significance.” In this usage, the presenter is referring to the fact that the P value is close to, but not actually less than, .05 (for a more thorough discussion on P values, please see Hall et al).1  This usage is probably ill-advised as it could overstate results that may not, in effect, indicate a difference between 2 study groups. So, should I say, “More third-year pediatric residents applied to a PHM fellowship compared to all other fellowships, which trended towards significance.” The seasoned listener will realize that there was, in fact, no significant difference between these groups, although the speaker would very much like there to be one. A more accurate usage of the word trend is in time series analyses. A time series analysis is based in regression analysis (essentially finding the “best fit” of a line to data) and is used to determine how an ordered independent variable (eg, time) is related to a dependent variable. For example, as many program directors could attest to, the number of applications for PHM fellowships has trended up over time.

Significant: This is a term that is both used and misused frequently. In common parlance, it is used to denote a notable difference between 2 groups. For example, I may comment “My dachshund is significantly grumpier than my beagle,” and the listener now knows that I have a grumpy dachshund as a pet, without giving much thought to the statistical significance of the difference in grumpiness between the 2 dogs. However, if I state, “Using the dog grumpiness scale, my dachshund has a significantly higher score than my beagle (P < .05).” With this statement, the listener now knows that using an a priori threshold of statistical significance (ie, .05), the probability that the observed difference in grumpiness scores being explained by random chance, instead of a true difference in grumpiness, is so small that we should reasonably conclude that dachshunds are grumpier than beagles. The confusion arises when the term “significant” is used without the supporting data. This would occur in the following sentence “Dachshunds are significantly grumpier than any other dog breed” without any data or analysis to support that statement. Without data, we can’t be sure if the author intends to communicate that dachshunds are notably grumpier or statistically grumpier than any other dog breed. Although using a qualifier such as “approaching significance” or “moderately significant” is fine when describing an everyday scenario, this type of usage is highly discouraged in scientific writing as it can misrepresent results that did not meet a statistical threshold.

Using these terms outside of these precise situations could cause confusion, whether the term is used while writing a scientific paper or in casual conversation. Use in casual conversation is likely to continue, though I acknowledge in some circles you may get some raised eyebrows if you state confidently that “bronchiolitis this year is significantly more severe than it was last year.”

Happy conferencing!


Your friendly methodologist

We thank Troy Hall, who provided methodological oversight to this manuscript.

FUNDING: No external funding.

CONFLICT OF INTEREST DISCLOSURES: The authors have indicated they have no potential conflicts of interest to disclose.

Drs Forster and Schondelmeyer were responsible for the concept behind this manuscript, drafting the manuscript, providing critical review of the manuscript, and approving the final draft.

Detecting health care disparities and the problem with P <0.05
Hosp Pediatr