METHOD GUIDE
Survival Analysis in SPSS
Time, Censoring and the Kaplan-Meier and Cox Tables
A survival model is only as sound as the Time and Status variables that feed it. Censoring codes, the choice of comparison test and the wording of each hazard ratio leave traces in the SPSS output that a reviewer will check.
IN SHORT
What the Two SPSS Survival Procedures Estimate
Survival methods analyze the time until an event such as death, relapse or dropout while keeping the information from participants who had not yet had the event when follow-up ended, who are called censored. The Kaplan-Meier procedure estimates the survival function and compares groups, and Cox regression estimates hazard ratios for one or more predictors. To run survival analysis in SPSS, open Analyze, Survival, then Kaplan-Meier or Cox Regression, enter a Time variable and a Status variable, and use Define Event to say which value means the event occurred (menus differ slightly by version). OnlineSPSS, a specialist statistical consultancy for academic, clinical and professional research, carries out and reviews this analysis through an in-house PhD-level statistician.
ANALYSIS SUPPORT
Analysis Support for This Method
Describing and comparing groups without adjustment is one stage, covered by Kaplan-Meier Analysis, and modeling covariates is a second stage with its own requirements.
Survival Analysis
The follow-up analysis planned within your study design, with the definition of time zero, the handling of censoring, the model and the reporting agreed for your data.
Cox Regression
Hazard ratios for one or several predictors, with the check on constant hazard ratios, the number of events behind the model and the intervals reported.
BEFORE YOU RUN IT
Building the Time and Status Variables
SPSS cannot tell when follow-up began or why someone left, so the Time and Status variables are built from the study design and the follow-up records.
Where the Clock Starts Time is a computed duration from a clear starting point, such as randomization or diagnosis, to the event or last contact. A time built from dates is checked for negative and impossible values.
Status Coding Status marks the event with one value, and everyone else is censored. Define Event must name that value, because a wrong code treats events as censored and drops the very outcomes of interest.
Reasons for Censoring The end of the study is usually harmless, but loss to follow-up linked to prognosis is not. The methods assume censoring is unrelated to the risk of the event, so how people left is described.
Events Behind the Model Precision and stability depend on the number of events, not the number of participants. Few events per covariate give unstable hazard ratios, and a competing event needs a different approach.
IN SPSS
How to Run Survival Analysis in SPSS
Both procedures sit under the Survival menu of IBM SPSS and share the Time, Status and Define Event setup, with labels that shift slightly across versions.
- Compute the follow-up time from the start date to the event or last contact, and create a Status variable.
- For a group comparison choose Analyze, Survival, Kaplan-Meier, place the follow-up time under Time and the status under Status, then use Define Event.
- Put the grouping variable under Factor, choose a test under Compare Factor, and request the survival table, mean and median survival and the survival plot under Options.
- For covariates choose Analyze, Survival, Cox Regression, then enter Time, Status, Define Event and the covariates.
- Declare categorical covariates under Categorical with a stated reference, and request CI for Exp(B) under Options.
- Read the Case Processing Summary first, then the survival and comparison tables, or Omnibus Tests and Variables in the Equation for Cox.
READING THE OUTPUT
Reading the Kaplan-Meier and Cox Tables in Order
Kaplan-Meier in SPSS prints its tables from one procedure and the Cox model from another, so each block below follows its own reading order.
| Output element | What it tells you | What to check |
|---|---|---|
| Case Processing Summary | The total N, the number of events and the censored N with its percent, by group in a comparison | That the event count matches the study records and the censored share is explained |
| Survival Table | The cumulative survival estimate at each event time, with its standard error and the number remaining | That late estimates are read with the number remaining, since they rest on few people |
| Means and Medians for Survival Time | Estimates with standard errors and 95% confidence intervals | That the median is reported with its interval, and the mean is read with care because it is limited to the longest time when that time is censored |
| Overall Comparisons | Log Rank (Mantel-Cox), Breslow and Tarone-Ware chi-square tests, with df and Sig. | That the test was chosen before looking at the results and is read beside the plotted groups |
| Survival Functions plot | The estimated survival by group as a step plot, with marks for censored cases | That groups whose lines cross are not summarized by one test alone, and the far right of the plot is read cautiously |
| Omnibus Tests of Model Coefficients (Cox) | Chi-square tests of the covariates together, with the -2 Log Likelihood | That the number of events behind the model is reported with it |
| Variables in the Equation (Cox) | B, SE, Wald, Sig., Exp(B) and the 95.0% CI for Exp(B) for each covariate | That Exp(B) is written as a hazard ratio per unit or against the reference group, with its interval |
| Log minus log plot or Cox w/ Time-Dep Cov | Whether the hazard ratio stays constant over follow-up | That roughly parallel lines, or a covariate-by-time term with no clear signal, are treated as support and not as proof |
REPORTING
Writing Up Follow-Up, Censoring and the Hazard Ratio
Counts of events and censored cases lead the write-up, ahead of any median or hazard ratio.
State the event, time zero and the end of follow-up, then give the number of participants, events and censored cases in each group, and the median follow-up with the method used to calculate it. For a Kaplan-Meier comparison write the median with its 95% CI [___, ___] where it was reached, followed by the test as χ²(df = ___) = ___, p = ___, naming the test. For the Cox model report HR = ___, 95% CI [___, ___], p = ___ for each covariate, the reference group and the number of events, and say how constant hazard ratios were assessed. The Kaplan-Meier procedure prints no interval for a difference in survival at a chosen time, so that comes from another program. Describe hazard ratios as differences in the rate of the event, and avoid causal wording for observational follow-up.
COMMON MISTAKES
Coding and Reading Errors in Survival Output
The first two concern what was counted as an event and when the clock started, and the rest concern plots and hazard ratios read for more than they show.
Censored Cases Dropped Removing participants who had no event, or counting them as failures, biases the estimates. Censored cases stay in with their follow-up time and a status that marks no event.
Time Measured Inconsistently Mixing months and days, or starting the clock at different points for different people, makes durations incomparable. One origin and one unit are fixed for everyone before the model runs.
Odds Language for Hazards Exp(B) in the Cox table is a ratio of event rates at a given moment. Describing it as odds of the event, or as cumulative risk being that many times higher, misstates what it measures.
Test Chosen After the Plot Switching among the Log Rank, Breslow and Tarone-Ware tests after seeing which p-value is smallest is selective reporting. The test is fixed in advance and named in the text.
Covariates Beyond the Events A Cox model with many covariates and few events gives unstable hazard ratios with wide intervals. The number of events, not the number of participants, limits how many covariates can enter.
Sparse Tail Over-Read The right-hand end of a Kaplan-Meier plot rests on few remaining participants, so its steps are noisy. The number remaining is read from the Survival Table and conclusions avoid that tail.
Tables already produced? Independent Statistical Review checks the coding of time and status, the test reported and the hazard ratio wording before you submit.
FREQUENTLY ASKED QUESTIONS
Questions About Survival Output in SPSS
Before any hazard ratio is interpreted, the event code and the choice of comparison test have to be right, and the answers start there.
How do I tell SPSS which value of my status variable is the event?
Enter the Status variable in the dialog, choose Define Event and type the value that marks the event, as a single value or a range. Every other value is treated as censored. Check the Case Processing Summary straight away: if almost everyone appears censored, or the event count is far from your records, the event code was entered the wrong way around.
Which test in the Overall Comparisons table should I report?
Report the test chosen before the data were examined, usually Log Rank (Mantel-Cox), and name it in the text. That test weights all event times equally and is most powerful when the hazard ratio is constant over time. Breslow gives more weight to early events, and Tarone-Ware sits between the two. Looking at all three and reporting the smallest p-value is selective reporting.
What does Exp(B) mean when I run Cox regression in SPSS?
Exp(B) is the hazard ratio: the multiplicative difference in the rate of the event at any moment for a one-unit increase in the covariate, or against the reference category. A value below 1 indicates a lower rate and above 1 a higher rate, with the interval in the 95.0% CI for Exp(B) columns when requested under Options. It is not a ratio of survival times or of cumulative risk, and it assumes the ratio is constant over follow-up.
Can OnlineSPSS check how I coded time and censoring?
Yes. An in-house PhD-level statistician can review how time zero, the event and censoring were defined, check the test and hazard ratios you report, or carry out the analysis on your data from the start. With the analysis settled, the survival tables and their text fall under Results Reporting & Interpretation. How the findings are interpreted remains your decision.
How do you protect my research data and confidentiality?
Your data and project details are handled confidentially, and project information is not published. Researcher names and institutions are not used publicly without permission. An NDA is available where required and can be arranged before sensitive materials are shared. Read more about confidentiality and data security.
What happens when I request an assessment, and how long does it take?
You share your research question, design and relevant materials, such as the dataset or the output you have, and a statistically trained Research Coordinator reviews the request and responds the same day. The analysis, deliverables, timeline and one fixed project fee are then confirmed before any work begins, so the timeline follows from the agreed scope.
Still have a question? Find more answers in our full FAQ.
NEXT STEP
Do Your Time and
Status Variables Hold Up?
Describe how time zero, the event and censoring were defined, with the number of events and any tables so far. Those details shape the scope that the assessment confirms.
No Commitment Required
Strictly Confidential
Fixed Project Fee
In-House PhD-Level Statisticians
Same-Day Response
Serving Researchers Since 2014