Logistic regression models the probability of a binary outcome, so its coefficients are read as odds ratios rather than straight-line slopes. An odds ratio above 1 means the predictor raises the odds of the event, a value below 1 means it lowers them, and a value of 1 means it makes no difference. Reading the output for your dissertation is mostly a matter of translating these odds ratios, their confidence intervals, and their p-values back into plain claims about your outcome.
Why logistic output is read differently from linear output
Because the outcome is a category and not a continuous measure, the model is fitted on the log-odds scale, where each coefficient is additive. Software then exponentiates the coefficient to give an odds ratio, which is the figure you actually interpret. This is the key break from the assumptions behind linear regression: there is no requirement of normally distributed residuals or constant variance, but you still need a reasonably large sample, little multicollinearity, and a roughly linear relationship between continuous predictors and the log-odds. Mistaking a log-odds coefficient for an odds ratio is one of the most common reporting errors your committee catches.
Reading the odds ratio and its confidence interval
Start with the odds ratio for each predictor. A value of 1.5 for a one-unit increase means the odds of the event are 1.5 times as high, or 50 percent higher, holding the other predictors constant. To judge whether that estimate is trustworthy, read its 95 percent confidence interval: if the interval excludes 1, the effect is statistically significant at the conventional level; if it spans 1, you cannot rule out no effect. A wide interval signals an imprecise estimate, often from a small sample or a rare outcome. These numbers are usually laid out in the same coefficients block you learn to navigate in how to interpret SPSS output.
Judging overall fit and significance
Individual predictors are not the whole story; you also report whether the model as a whole is significant. The omnibus test of model coefficients, a likelihood-ratio chi-square, tells you if your predictors together improve on the null model. The Hosmer-Lemeshow test checks calibration, and pseudo R-squared measures such as Nagelkerke give a rough sense of explained variation. Classification accuracy and the area under the ROC curve describe how well the model separates the two outcome groups. Report these alongside the coefficient table so the reader can see both the overall performance and the individual effects.
Interpreting odds ratios as percentage change
Examiners often want the odds ratio translated into a plain percentage, because that is the form a non-specialist reader grasps instantly. The rule is simple: subtract 1 from the odds ratio and multiply by 100. An odds ratio of 1.25 becomes a 25 percent increase in the odds; an odds ratio of 0.80 becomes a 20 percent decrease. Two cautions keep this honest. First, an odds ratio describes a change in odds, not in raw probability, and the two diverge sharply when the outcome is common rather than rare. Second, for a continuous predictor the figure applies to a one-unit change, so a small unit such as a single point on a scale can hide a large effect across the full range. When in doubt, report the odds ratio for a meaningful increment and state the unit explicitly so the practical size of the effect is unmistakable.
Logistic regression vs linear regression
The choice between the two methods follows entirely from your outcome variable. Use linear regression when the outcome is continuous, such as a test score or a measurement, because it models a straight-line change in the mean. Use logistic regression when the outcome is binary, such as pass or fail, because a linear model would predict impossible probabilities below 0 or above 1 and would violate its own assumptions. Beyond the outcome, the differences cascade: linear regression yields slopes you read directly, while logistic regression yields odds ratios after exponentiation; linear regression leans on normal residuals and constant variance, while logistic regression drops both and instead asks for a large enough sample and a roughly linear link to the log-odds. Picking the wrong family is a setup error no amount of careful interpretation can repair, which is why a quick check with the statistical test selector is worthwhile before you model.
Reading the logistic regression output table
Most software lays the results out in a single coefficients block, and knowing the columns saves hours. The B column holds the raw log-odds coefficient, which is rarely interpreted on its own; the S.E. column is its standard error; the Wald statistic and its significance value test whether that predictor differs from zero. The column you act on is Exp(B), the exponentiated coefficient, which is the odds ratio itself, usually flanked by its lower and upper 95 percent confidence bounds. Read across each row rather than down each column, so every predictor is judged as a complete picture of direction, size, and certainty. The wider habit of moving methodically through a results pane is the same one built in reading SPSS output in a fixed order, and it keeps you from quoting a log-odds figure where an odds ratio belongs.
Writing the result up cleanly
Present each predictor with its odds ratio, confidence interval, and exact p-value, then translate the headline ones into ordinary language so a non-statistician on your panel can follow them. Follow the conventions in the APA structure for a statistical result for decimals and italics. If your design also involves an indirect or conditional effect on a binary outcome, the same odds-scale reading carries into mediation and moderation analysis, where interactions are interpreted on the odds metric rather than the raw probability.