HLT-362V Correlation and Causation: Guide, Example & Answers

Two things can move together in your data without one actually causing the other. That single idea trips up more HLT-362V students than almost any other statistics topic, because the math behind correlation is simple, but the reasoning behind causation is not.

This guide walks through what correlation and causation actually mean, why a statistically significant relationship is not automatically a cause-and-effect relationship, and how to spot the difference in a published healthcare study. It uses a real vitamin D and COVID-19 example that shows exactly how researchers moved from an observed association to a more careful causal conclusion, plus a practical checklist you can use on your own HLT-362V article analysis or discussion post.

GCU’s current course description frames HLT-362V around understanding the appropriateness of statistical methods used in published healthcare research, and this distinction sits right at the center of that goal. Assignment requirements can vary by current course materials and instructor instructions, so always verify the current rubric and assignment directions in your GCU course before you submit.

The central question to keep coming back to throughout this guide: does this statistical relationship actually show that one variable caused the other, or just that the two tend to move together?

This assignment is part of HLT-362V Applied Statistics for Health Care at GCU. For a complete guide to every HLT-362V assignment in sequence, see the HLT-362V Applied Statistics for Health Care guide.

HLT-362V Correlation and Causation at a Glance

If you are short on time, start with this table.

Correlation does not establish causation. That said, it would be an oversimplification to say observational studies can never provide causal evidence. Strong causal inference can sometimes be built from observational data using appropriate study designs, adjustment methods, and assumptions. Randomized experiments generally provide stronger causal evidence when they are ethical and feasible, but “observational” and “useless for causal reasoning” are not the same thing.

What Is Correlation?

Correlation describes a statistical relationship between two variables: whether, and how strongly, they tend to change together.

Positive correlation

As one variable increases, the other tends to increase as well. Example: as nursing shift length increases, self-reported fatigue scores tend to increase.

Negative correlation

As one variable increases, the other tends to decrease. Example: as patient education sessions increase, hospital readmission rates tend to decrease.

Little or no linear correlation

There is little evidence of a linear relationship between the two variables. Example: shoe size and blood glucose level would be expected to show little to no correlation.

Pearson’s correlation coefficient (r) describes the direction and strength of a linear relationship, ranging from −1 to +1:

• r close to +1: strong positive relationship
• r close to −1: strong negative relationship
• r close to 0: little to no linear relationship

The important thing to hold onto here: correlation measures a relationship. It does not measure, describe, or prove a cause-and-effect mechanism. A high r value tells you two variables move together consistently; it tells you nothing about why.

Correlation Visualized

Picture three scatterplots side by side:

• Positive correlation: points trend upward from left to right, clustering tightly around an upward line.
• Negative correlation: points trend downward from left to right.
• Little/no correlation: points scattered with no clear pattern.

Now picture changing a handful of data points in the positive-correlation scatterplot so they cluster even more tightly around the line. The r value climbs closer to +1. What does not change is the underlying question of why those two variables move together. A tighter scatterplot is stronger statistical evidence of association. It is not, by itself, stronger evidence of cause and effect. That distinction is the entire point of this guide, and it holds true no matter how strong the correlation looks on the page.

What Is Causation?

Causation means that changing one variable (the exposure or intervention) produces a change in another variable (the outcome), under clearly defined conditions.

Healthcare examples of causal relationships:

• A medication causing a measurable physiological effect, such as a beta-blocker lowering heart rate.
• A structured mobility intervention reducing post-surgical recovery time.
• Smoking increasing the risk of certain cancers, established through decades of converging evidence from multiple study designs.

Causal questions are different from association questions. Compare:

• Association question: “Are smoking and lung cancer related?”
• Causal question: “Does smoking increase the risk of lung cancer?”

The first question can be answered with a single observational dataset. The second requires much more: a clear temporal sequence, ruling out alternative explanations, and ideally converging evidence from multiple types of studies.

What Is the Difference Between Correlation and Causation?

Correlation can be a legitimate reason to investigate further. It is evidence that motivates a research question. It is not, by itself, sufficient to answer a causal one.

What Separates Correlation From Causation?

The key issue is not simply whether two variables are related. Before treating a relationship as causal, researchers (and HLT-362V students evaluating a study) need to ask:

1. Could a third variable explain the relationship?
2. Could reverse causation explain the relationship?
3. Was the exposure or intervention assigned by the researcher, or only observed?
4. Were participants randomized?
5. Was there a control or comparison group?
6. Were important confounders measured and statistically addressed?
7. Is the temporal sequence clear (did the exposure clearly come before the outcome)?
8. Is there a plausible biological or clinical mechanism?
9. Do multiple studies using different designs produce consistent findings?
10. Does the totality of evidence support a causal interpretation, or only an association?

Causal inference is based on the weight of evidence and study design taken together. It is never established by a correlation coefficient alone.

Confounding Variables

A confounder is a variable that is related to both the exposure and the outcome and can distort the apparent relationship between them.

Healthcare example: A researcher observes that patients who exercise more tend to have better cardiovascular outcomes. It’s tempting to conclude that exercise causes better cardiovascular outcomes. But age is related to both: younger patients tend to exercise more, and younger patients also tend to have better cardiovascular outcomes independent of exercise habits.

Simple diagram of the relationship:

Exercise  <---  Age  --->  Cardiovascular outcome

Age influences both exercise habits and cardiovascular outcomes, which means it can make the exercise-outcome relationship look stronger, weaker, or different than the true underlying relationship. If a study does not account for age, the exercise-outcomes association it reports may be partly, or even mostly, explained by age rather than exercise itself.

Confounding is not the same as mediation. A confounder is a separate cause of both the exposure and the outcome that sits outside the causal pathway you’re trying to study. A mediator, by contrast, sits on the causal pathway between exposure and outcome (for example, if exercise improves cardiovascular outcomes partly through reducing blood pressure, blood pressure is a mediator, not a confounder). HLT-362V students most often need to recognize confounding, so that’s the concept to prioritize.

Reverse Causation

Reverse causation occurs when the outcome influences the exposure, rather than the exposure causing the outcome, which is the opposite of what a researcher initially assumed.

Healthcare example: A researcher observes that patients with low physical activity levels tend to have poorer overall health and concludes, “Low physical activity causes poor health.” But the reverse may also be true: poor health (chronic pain, fatigue, cardiopulmonary limitations) may cause people to become less physically active in the first place. The observed association could reflect either direction, or both simultaneously.

This is exactly why temporal sequence matters so much when evaluating a causal claim. If a study measures exposure and outcome at the same point in time (a cross-sectional design), it usually cannot distinguish “activity level affects health” from “health affects activity level.” Longitudinal designs, where exposure is measured clearly before the outcome, help address this problem, though they do not eliminate every alternative explanation on their own.

Correlation Does Not Equal Causation

A statistically significant correlation does not automatically mean that one variable caused the other.

Even a very strong correlation can be misleading if:

• A confounder exists that hasn’t been measured or adjusted for
• Reverse causation is possible
• The sample is biased or unrepresentative
• The variables are measured poorly or inconsistently
• The relationship is coincidental
• The study design cannot establish a clear temporal sequence

The correct framing is not “correlation never means causation.” It is: correlation alone does not establish causation. A correlation can absolutely be part of a larger body of evidence that eventually supports a causal conclusion. It just cannot carry that conclusion by itself.

Healthcare Examples of Correlation Without Causation

Example 1: Ice cream sales and drowning incidents

Both ice cream sales and drowning incidents rise during the summer months. Temperature (and the outdoor swimming activity that comes with it) is the third variable driving both. Ice cream does not cause drowning; both are downstream of warm weather and summer activity patterns.

Example 2: Hospital staffing and adverse events

Hospitals with more nurses on a given unit sometimes also report more documented adverse events. It would be a mistake to conclude that higher nurse staffing causes adverse events. A more plausible explanation is that hospitals often increase staffing in response to higher patient acuity, and higher patient acuity independently increases the risk of adverse outcomes. Staffing level and adverse events may both be responding to the same underlying driver: how sick the patients are.

Example 3: Coffee consumption and heart disease

Observational studies linking coffee consumption to cardiovascular outcomes are vulnerable to multiple confounders: smoking, age, diet, sleep quality, and occupational stress can all be related to both coffee habits and heart disease risk. A raw association between coffee and heart disease, without adjusting for these factors, should not be read as “coffee affects heart disease risk” without a closer look at how the study handled confounding.

Example 4: Vitamin D and COVID-19 outcomes

Early observational studies found that patients with lower vitamin D status tended to have worse COVID-19 outcomes. That association was real and was reported consistently across many observational studies. It did not, on its own, establish that vitamin D deficiency caused more severe disease. As the worked example below shows, many of the same conditions associated with low vitamin D status (older age, obesity, chronic disease) are independently associated with worse COVID-19 outcomes, which makes this a textbook confounding problem (Thacher, 2022).

Need help with your HLT-362V Correlation and Causation assignment? Send Gradevia a message on WhatsApp for a quick scope review if you’re not sure how to apply these ideas to the specific article your instructor assigned.

HLT-362V Correlation and Causation Worked Example: Vitamin D and COVID-19

Step 1: Identify the variables

Exposure: vitamin D status (typically measured as serum 25-hydroxyvitamin D, or 25(OH)D) Outcome: COVID-19 severity and clinical outcomes (hospitalization, ICU admission, mortality)

Step 2: Identify the observed correlation

Multiple observational studies conducted early in the pandemic reported that patients with lower vitamin D status tended to have worse COVID-19 outcomes, including higher rates of severe disease and mortality (Thacher, 2022).

Step 3: Ask whether correlation proves causation

No. An observed association between vitamin D status and COVID-19 severity in observational data does not, by itself, establish that low vitamin D caused the worse outcomes.

Step 4: Identify possible confounders

Thacher (2022) specifically identifies high-risk conditions that are independently associated with both lower vitamin D status and worse COVID-19 outcomes, including:

• Age over 65
• Obesity
• Chronic kidney disease
• Diabetes
• Cardiovascular disease
• Chronic lung disease
• Certain cancers and mental health conditions
• Limited time spent outdoors / reduced sun exposure

Any of these could independently drive both lower vitamin D levels and worse COVID-19 outcomes, which means an unadjusted association between the two can look stronger than the true causal effect, if any exists.

Step 5: Consider reverse causation

It’s also plausible that serious illness itself affects vitamin D status rather than the other way around: acutely ill, hospitalized patients often have reduced sun exposure, altered nutritional intake, and inflammation-related changes in vitamin D metabolism. This means the direction of the relationship is not automatically obvious just from an association.

Step 6: Evaluate the study design

Most of the early clinical research on vitamin D and COVID-19 was observational, which is the design most vulnerable to confounding of the type described above (Thacher, 2022). Observational data can adjust statistically for measured confounders, but only for the confounders researchers actually measured and included in their models. Unmeasured confounders remain a limitation.

Step 7: Identify stronger evidence

A randomized controlled trial addresses this problem directly by assigning vitamin D supplementation versus placebo, so that confounding factors are, in principle, distributed evenly across both groups. One such RCT, conducted in ICU patients with severe COVID-19 and confirmed vitamin D deficiency, found no significant benefit of vitamin D supplementation across the outcomes measured, including duration of mechanical ventilation (Domazet Bugarin et al., 2023).

Step 8: State the correct conclusion

The accurate conclusion distinguishes two very different statements:

• “Vitamin D status was associated with COVID-19 outcomes in observational studies.” This is supported by the evidence.
• “Vitamin D deficiency caused severe COVID-19.” This is a stronger causal claim that the observational evidence alone cannot support, and that at least one RCT did not confirm when it directly tested supplementation as an intervention (Domazet Bugarin et al., 2023).

This is exactly the kind of distinction HLT-362V article-analysis assignments are testing for: can you tell the difference between what a study actually showed and a causal claim that goes beyond what the data can support?

Stuck applying this kind of analysis to your own assigned research article? Message Gradevia on WhatsApp and describe the study you’re working with.

How Randomized Controlled Trials Help Establish Causation

In a randomized controlled trial (RCT), participants are randomly assigned to a treatment group or a control group, rather than the exposure being observed as it naturally occurs.

Why randomization strengthens causal inference: random assignment is designed to balance both measured and unmeasured characteristics between groups, on average. If age, comorbidities, baseline health status, and dozens of other factors are distributed roughly evenly between the treatment and control groups purely by chance, then a difference in outcomes between the groups becomes much more plausibly attributable to the intervention itself, rather than to a confounder.

Important nuance: randomization does not guarantee that groups are identical. It is designed to balance confounding factors on average across the study, particularly in adequately sized trials, but chance imbalances can still occur, especially in smaller trials. Randomization strengthens causal inference; it does not eliminate every possible source of bias.

Why RCTs aren’t always possible:

• Ethical limitations: it would be unethical to randomly assign people to smoke, or to withhold a known-effective treatment from a control group.
• Cost and feasibility: RCTs are often expensive and logistically demanding.
• Long latency periods: some outcomes (certain cancers, for example) take years or decades to develop, making trials impractical.
• Rare exposures or outcomes: if very few people are exposed, or the outcome is very rare, a trial may not be feasible to power adequately.

When RCTs are not possible, well-designed observational causal-inference methods (such as propensity score matching, instrumental variable analysis, or natural experiments) can still contribute meaningful evidence toward a causal question, even though they generally carry more assumptions than a randomized design (Fernainy et al., 2024).

Observational vs Experimental Studies

For HLT-362V purposes, the practical takeaway is this: when you read a published study, check whether the researchers assigned the exposure or only observed it. That single detail tells you a great deal about how much weight the study’s causal language deserves.

How to Identify Correlation vs Causation in a Research Article

Use this checklist when you’re reading a study for an HLT-362V assignment:

1. What are the two variables being studied?
2. Are they associated, and how strongly?
3. How were the variables measured?
4. Was the study observational or experimental?
5. Was the exposure assigned by the researchers, or only observed?
6. Was randomization used?
7. Could a confounding variable explain the relationship?
8. Could reverse causation occur?
9. Is there a clear temporal relationship (exposure before outcome)?
10. Does the author claim causation, or only association?
11. Does the study design actually justify that claim?
12. Are the findings statistically significant?
13. Are the findings clinically meaningful, separate from statistical significance?
14. Are alternative explanations addressed or ruled out?

If you can answer most of these questions about a study, you’re in a strong position to evaluate whether its conclusions are appropriately worded, or whether the authors (or a secondary source reporting on the study) overstated a causal claim that the design doesn’t fully support.

Common Mistakes in HLT-362V Correlation and Causation

1. Saying correlation proves causation.
2. Assuming statistical significance proves causation.
3. Ignoring confounding variables entirely.
4. Ignoring the possibility of reverse causation.
5. Assuming a large sample size automatically proves causation.
6. Treating an observational study’s findings as if they came from an experiment.
7. Confusing an observed association with an explained biological mechanism.
8. Ignoring selection bias in how the sample was recruited.
9. Ignoring measurement problems (how the exposure or outcome was actually assessed).
10. Treating a single study as proof of a causal relationship.
11. Confusing statistical significance with clinical significance.
12. Claiming that observational studies can never contribute to causal inference (they can, under the right design and assumptions; they just require more caution than randomized evidence).

Quick Decision Tool: Can I Claim Causation?

Stronger evidence for causation

• Temporal sequence is clear (exposure precedes outcome)
• Randomization was used when feasible
• A comparison or control group exists
• Major confounders are identified and statistically addressed
• Results are consistent across studies and populations
• A plausible biological or clinical mechanism exists
• Findings have been replicated
• Alternative explanations have been considered and reasonably ruled out

Warning signs

• Observational design only, with no adjustment for likely confounders
• No comparison group
• Major known confounders not addressed
• Reverse causation is plausible and unaddressed
• Small or biased sample
• Exposure or outcome measured poorly or inconsistently
• Authors (or media coverage of the study) make causal claims based on correlation alone

HLT-362V Connection to Nursing Practice

Understanding correlation and causation is not just an HLT-362V exercise. It directly supports how nurses:

• Interpret evidence-based practice research before recommending a change in care
• Evaluate treatment claims made in professional literature, continuing education, or by patients and families
• Understand which risk factors are genuinely modifiable versus which ones are markers of something else going on
• Identify misleading health claims in media coverage or marketing of supplements, devices, and interventions
• Communicate research findings accurately to colleagues, patients, and families, without overstating what a study actually showed
• Support evidence-based patient education that reflects the real strength of the evidence, not an inflated version of it

A nurse who understands this distinction is better equipped to ask the right question when a new study, treatment claim, or risk factor makes headlines: is this actually causal, or just associated?

HLT-362V Correlation and Causation FAQ

Is 100% correlation causation? No. A 100% (or perfect) correlation still does not automatically establish causation. A perfect correlation means the two variables have a perfectly consistent statistical relationship within the data observed, but it does not by itself tell you which variable is causing which, whether a third variable explains both, or whether the relationship is coincidental. Ice cream sales and drowning incidents can show a strong, consistent seasonal correlation because temperature drives both, without ice cream having any causal role in drowning risk.

What is the difference between correlation and causation? Correlation describes a statistical relationship between two variables: they tend to move together. Causation means that changing one variable produces a change in the other. For example, patient satisfaction scores and nurse staffing ratios might be correlated, but only a study specifically designed to test causation, such as an intervention that changes staffing and then measures satisfaction, could support a causal claim.

What separates correlation from causation? The distinction depends on the strength and type of causal evidence available: temporal sequence, study design, randomization where appropriate, presence of a control or comparison group, how well confounders were assessed, whether alternative explanations were considered, and whether findings are consistent across multiple studies. Correlation is evidence of association. It is not, by itself, evidence of cause and effect.

Can you give me an example of a correlation that doesn’t imply causation? In healthcare: hospitals with higher nurse staffing levels sometimes also report more adverse events. It would be incorrect to conclude that higher staffing causes adverse events, because hospitals often increase staffing specifically when caring for sicker, higher-acuity patients, who independently carry a higher risk of adverse outcomes. Patient acuity is a likely confounder. A simpler, non-healthcare example is ice cream sales and drowning incidents, both of which rise in the summer because of a shared third variable: warm weather.

What is a confounding variable? A confounder is a variable related to both the exposure and the outcome that can distort, exaggerate, or mask the apparent relationship between them. Age, for example, often confounds relationships between health behaviors (like exercise) and health outcomes, since age is independently related to both.

Can correlation ever provide evidence for causation? Yes, as one piece of a larger body of evidence. A well-documented, consistent correlation across multiple studies and populations, combined with a clear temporal sequence, a plausible mechanism, and ideally some experimental support, can contribute to a causal conclusion. Correlation alone, from a single observational study, cannot.

Does a statistically significant correlation prove causation? No. Statistical significance tells you that an observed relationship is unlikely to be due to random chance alone. It says nothing about whether the relationship is causal, confounded, or the result of reverse causation.

What is reverse causation? Reverse causation occurs when the presumed outcome actually influences the presumed exposure, rather than the other way around. For example, a researcher might observe that low physical activity is associated with poor health and conclude that inactivity causes poor health, when in fact poor health may be causing people to become less active.

Why are randomized controlled trials useful for causal inference? Random assignment is designed to balance both known and unknown confounding factors between the treatment and comparison groups, on average. This makes it more plausible that a difference in outcomes reflects the effect of the intervention itself rather than a pre-existing difference between groups.

Can observational studies show causation? Observational studies can demonstrate association clearly, and under the right conditions (careful measurement of confounders, appropriate statistical adjustment, clear temporal sequence, and sometimes specialized causal-inference methods) they can contribute meaningful evidence toward a causal question. They are generally more vulnerable to confounding than randomized experiments, though, which is why RCT evidence is typically weighted more heavily when it’s available (Fernainy et al., 2024).

What is the difference between association and correlation? The terms are often used interchangeably in healthcare research, though “correlation” is sometimes used more specifically for a linear statistical relationship measured by a coefficient like Pearson’s r, while “association” is a broader term that can describe relationships between categorical variables too, such as those measured by chi-square.

Why is correlation and causation important in nursing? Because nurses constantly encounter research findings, treatment claims, and risk-factor information that require this exact distinction to interpret correctly. Overstating an association as causal can lead to misapplied interventions or misleading patient education; understating a well-supported causal relationship can delay adoption of effective practices.

How do I explain correlation and causation in an HLT-362V assignment? Identify the two variables, state clearly whether they are correlated or causally related based on the study design, and explicitly name at least one alternative explanation (confounding, reverse causation, chance, bias) if the study is observational. Instructors are generally looking for you to demonstrate that you understand why a relationship might not be causal, not just that you can recite the phrase “correlation does not equal causation.”

What should I look for when analyzing a healthcare research article? Start with the checklist in the “How to Identify Correlation vs Causation in a Research Article” section above: study design, whether exposure was assigned or observed, whether confounders were addressed, whether temporal sequence is clear, and whether the authors’ conclusions match the strength of evidence their design can actually support.

Related Content

HLT-362V ANOVA and Chi-Square Exercises: Examples, Answers, Solved Problems & Guide

NRS 425 Epidemiology and Community Health: GCU Guide

NRS 425 Health Promotion and Population Health: GCU Course Guide

NRS-465: Applied Evidence-Based Project and Practicum

References

Basu, A. (2021). Concept of confounding variables. Qeios. https://doi.org/10.32388/62T3J0

Domazet Bugarin, J., Dosenovic, S., Ilic, D., Delic, N., Saric, I., Ugrina, I., Stojanovic Stipic, S., Duplancic, B., & Saric, L. (2023). Vitamin D supplementation and clinical outcomes in severe COVID-19 patients: Randomized controlled trial. Nutrients, 15(5), 1234. https://doi.org/10.3390/nu15051234

Fernainy, P., Cohen, A. A., Murray, E., Losina, E., Lamontagne, F., & Sourial, N. (2024). Rethinking the pros and cons of randomized controlled trials and observational studies in the era of big data and advanced methods: A panel discussion. BMC Proceedings, 18(Suppl 1), 3. https://doi.org/10.1186/s12919-023-00285-8

Thacher, T. D. (2022). Evaluating the evidence in clinical studies of vitamin D in COVID-19. Nutrients, 14(3), 464. https://doi.org/10.3390/nu14030464

Grand Canyon University. (n.d.). HLT-362V: Applied Statistics for Health Care Professionals — course description. Course catalog, Grand Canyon University. (Course-level description reflected in this article; consult your current course syllabus for assignment-specific requirements, as these are set by individual instructors and may change by term.)