Most pipo no dey misread statistics because dem no sabi mathematics – na because nobodi ever explain to dem wetin p-value or confidence interval dey really promise. This article go waka you through five things wey go make you fit read any health or research finding well well.

Know which kind data you dey look At

Every variable for one study na either category or number. Blood type, whether person dey smoke, pain level (small, medium, serious) – all dis na categories. Height, age, blood pressure – na numbers.

Numbers dey split again into two kind: discrete (whole numbers wey you fit count, like how many times person go emergency room – you no fit get 2.5 times) and continuous (any value wey fit dey inside one range, like weight – 68.3kg dey perfectly make sense).

Dis one no be just grammar wey no get use. E dey decide which kind summary statistics and which kind chart dey correct to use. You no fit “average” pain level the way you dey average height for centimeter, because the “gap” between small pain and medium pain no be fixed equal unit like centimeter be.

Before you trust any claim wey dem talk about number for one study, ask yourself: which kind data bring dis number come?

Descriptive Statistics: Mean, Median, and the outlier wahala

Mean, median, and mode na three different way to answer “wetin be typical?” Mean na the mathematical average. Median na the number wey dey middle when you arrange everything. Mode na whichever number wey dey show up pass.

When data no get any extreme outlier, all three go roughly agree. Wahala go start when one or two extreme number show face.

Make we talk am with example: patients for waiting room dey 22, 24, 25, 26, and 27 years. Mean na 24.8, median na 25 – basically na the same story. Now make we change the last patient to 90 years old: 22, 24, 25, 26, 90. Mean jump go 37.4, even though nobody for that room dey “37-ish” at all. Median no move reach anywhere, e still dey 25 – wey be more honest description of wetin typical patient dey look like.

Na exactly why hospital bill, emergency room wait time, and income dem dey report as median, no be mean – because small number of extreme case fit drag mean comot from wetin dey really typical. Anytime you see average, ask yourself whether na small outlier dey do the dragging.

Wetin p-value dey really tell you

P-value dey answer only one small question: if treatment no dey do anything at all, how surprising dis kind data go be?

Make we yan am like coin-flip. If you flip coin 10 times, you get 9 heads – fair coin fit still do dat one, e just dey surprising. P-value dey measure exactly dat kind surprise, assuming say “null hypothesis” (say nothing dey really happen) na true.

P-value wey dey below 0.05 na just normal agreement wey pipo use for “e dey quite surprising” – e no be magic machine wey dey detect truth. Three myth wey we need commot for mind once and for all:

• “p < 0.05 mean say drug get 95% chance to work.” Na lie – p-value no dey talk anything about the probability wey hypothesis dey true. Na about how surprising the data go be if drug no dey do anything.
• “p > 0.05 prove say drug no get effect.” Na lie – no evidence no mean say evidence no dey (i.e., absence of evidence isn’t evidicence of absence). Fit be say the study small pass to detect real effect.
• “Statistically significant mean say e important for clinic.” Na lie – if sample big well well, even small effect wey no mean anything fit become statistically significant.

Confidence Interval: The Honest Range

Point estimate na your one best guess. Confidence interval na the honest range of guesses wey fit still dey correct, based on the data wey you get.

Picture am like fishing net instead of single hook. 95% confidence interval mean say: if you repeat the study 100 times, and you throway net like dis every time, about 95 out of dat 100 net go catch the true value. E no mean say “95% chance say true value dey inside dis particular range” – dat 95% dey describe how reliable the method be for long run, e no dey describe dis one interval wey you dey look.

Two things wey you must check anytime you see confidence interval:

• How wide e be – narrow interval mean say the estimate precise and you fit trust am. Wide one mean say plenty uncertainty still dey, even if the average number sound impressive.
• E dey cross the “no effect” line? – zero, for difference, or one, for ratio like relative risk. If e cross dat line, the result no be statistically significant, no matter how the point estimate promise well well.

Two trials fit report the exact same average effect, but dem go tell completely different story once you check their interval – one precise and clearly real, the other one too uncertain make you rule out “no effect at all.”

How to Read Results Section Like Pro, and Yan Am for Simple Language

Put the first four idea together, you go get simple checklist wey you fit use read any results section:

1. Who dem study, and how many? Result wey come from 40 pipo for one hospital no go generalize the same way as result wey come from 4,000 pipo across 20 sites.
2. Wetin be the effect size, no be just p-value? “Significant” no dey tell you whether the effect big reach to matter.
3. How wide the confidence interval be? Precision matter well well, same way as the main number.
4. Statistical significance mean clinical significance for dis case? Ask am directly – even real effect fit too small to change any decision.

The last step na the one wey pipo dey skip: to yan am for simple language without losing the honest uncertainty. Results section fit report say “mean reduction for systolic blood pressure na 8 mmHg (95% CI: 5–11 mmHg; p < 0.001).” If you translate am for patient, e go sound like: “On average, dis treatment reduce blood pressure by about 8 points, and we fairly confident say the true effect dey somewhere between 5 to 11 points – e no be just luck.”

Na the same fact. Na the same honesty about uncertainty. Zero grammar wahala. Na dat one be the real skill – no be just to understand finding, but to fit hand am over to another person complete, as e be.

You Need Help With Your Own Data?

oores Analytics® dey offer affordable biostatistics consulting, data analytics, and hands-on R training – and first consultation no go cost you anything. Waka reach us or send message go info@ooresanalytics.com make we discuss your research.