DBA · DATA ANALYSIS · 04
0%
DATA ANALYSIS FOR THE DBA · PART 4 OF 6

Reading the Output

Software hands you a table of numbers. This page turns each one — coefficient, p-value, confidence interval, R², odds ratio — into a sentence you can defend.

1 · THE COEFFICIENT TABLE
LINEAR REGRESSION OUTPUT

Every regression prints this table

Here is a real-shaped output: we explain a satisfaction score (0–100) from three drivers. Tap any row and the panel translates that row into plain English. Tap the fit chips below too.

Interactive 1 — tap each row and each chip.
VariableCoefficientStd. errorp-value95% CI
Perceived price−4.201.050.001[−6.3, −2.1]
Wait time−2.800.950.004[−4.7, −0.9]
Loyalty member+0.900.720.210[−0.5, +2.3]
R² = 0.62Adj. R² = 0.60n = 240F-test p < 0.001
Tap a row or chip…
2 · WHAT A p-VALUE REALLY IS
THE MOST MISUNDERSTOOD NUMBER IN RESEARCH

The logic of "could this be a fluke?"

Start from a sceptical position called the null hypothesis: "this variable has no real effect; its true coefficient is 0." Your data shows some effect — but random sampling alone produces apparent effects even when none exist. The p-value answers one precise question:

p = P( seeing an effect this big or bigger | the truth is zero )
small pthe data would be very surprising if the effect were really zero → evidence the effect is real. By convention p < 0.05 is "statistically significant."
large pthis much apparent effect happens easily by chance → you cannot rule out zero.

Below: the bell curve is what the coefficient would look like across many samples if the true effect were zero. Slide your observed coefficient outward. The shaded tails are the p-value — the chance of landing that far out by luck alone.

Interactive 2 — push the observed effect away from zero; watch p shrink.
p-value (two-tailed)
Verdict at α=0.05
What p is NOT. It is not the probability the hypothesis is true. It is not the size or importance of the effect (a tiny, useless effect can be "significant" with a big enough sample). And p > 0.05 does not prove "no effect" — only that you couldn't detect one. Report effect sizes and CIs, never p-values alone.
3 · CONFIDENCE INTERVALS
THE NUMBER EXAMINERS PREFER TO p

A range of plausible truths

A coefficient of −4.2 is your single best estimate, but it has uncertainty. The 95% confidence interval [−6.3, −2.1] gives the range of values consistent with your data. The interpretation that matters:

Slide the sample size and watch the CI tighten around the estimate — and watch a borderline effect cross from "not significant" to "significant" purely because precision improved.

Interactive 3 — more data narrows the interval.
Estimate
−2.0
95% CI
Significant?
4 · ODDS RATIOS (LOGISTIC OUTPUT)
A DIFFERENT TABLE, A DIFFERENT RULE

Logistic regression reports odds ratios

Because logistic models a probability through the log-odds, its coefficients are exponentiated into odds ratios (OR). The single rule:

OR > 1 raises the odds of "yes" · OR < 1 lowers them · OR = 1 means no effect. A CI for an OR signals significance when it excludes 1 (not 0 — because 1 is the "no effect" point on the odds scale).

Tap each row of this churn model.

Interactive 4 — tap each row. Watch where the "no effect" line (OR=1) falls.
VariableOdds Ratiop-value95% CI for OR
Filed a complaint2.500.002[1.4, 4.5]
Tenure (per year)0.700.010[0.55, 0.90]
Newsletter member1.050.480[0.85, 1.30]
Tap a row…

Reading these two tables — coefficients with CIs, and odds ratios — is the bulk of what any examiner will ask you to interpret.

Scroll progress saves on this device. Part of the DBA Data-Analysis series by Jan Erik Meidell.
JMJan Erik Meidell