A Quick Note On Where These Numbers Come From
The passenger and crew data for the Titanic is publicly available, most commonly in the Kaggle Titanic dataset that contains 891 rows with variables like survival status, class, age, sex, fare, and port of embarkation. I've worked with this dataset enough times that the patterns are basically second nature to me now, but there are still people who treat it like a beginner's toy without understanding where the gaps in the data actually are. The raw survival count is 549 people out of 2,224 total souls on board. That works out to a 31.3% overall survival rate. But numbers like that are almost useless on their own because they hide the most important story. What actually matters is how survival probability shifted depending on which bucket you fell into when the ship went down. First class had a survival rate around 62-63%. Second class came in at roughly 41%. Third class was about 25%. The deck crew and other crew members who make up the rest of the passenger list sat somewhere near 24% survival. This pattern alone tells you that socioeconomic position was a significant factor, but it's not the whole picture and anyone who stops there is missing the deeper structure.
When I first started pulling this data years ago, I made the mistake of treating class as purely a proxy for money. It's not. Class on the Titanic also mapped directly to proximity to the lifeboats, language barriers, and in some cases physical access through the ship's layout. The staircase positions and door placements literally determined who could move fast enough to reach the boat deck.
Gender Was The Strongest Predictor
Women survived at roughly 74% overall. Men survived at about 19%. This is one of the most consistent signals in the entire dataset and it holds across every class level, though the gap is widest in first class where women hit nearly 97% survival while men were closer to 33%. The woman-first protocol wasn't just a social convention. There are documented instances where men in formal evening wear were physically restrained from boarding lifeboats by officers who had strict orders. A lot of modern analyses treat this as a moral story. It's really just a stark example of how one binary variable can dominate a predictive model.
Get the Full Details
Age And Family Structure Matter More Than You'd Expect
Children under five had a survival rate above 60%. That sounds high until you remember the sample size was small. Children between five and twelve tracked around 53%. Adult survival dropped sharply from there unless you were a woman in first or second class. Passenger family size is another feature that quietly matters. Solo travelers had a lower survival rate than those traveling with one or two close family members. The effect gets messy past that because larger parties faced bottlenecks at exits and stairwells. I found this out the hard way when I tried to model SibSp and Parch separately at first and got garbage results. Merging them into a single family size variable fixed the noise and improved predictive accuracy noticeably.
Fare As A Signal
Higher fare correlates with higher survival, but the relationship isn't linear. Someone who paid a steep fare in third class like a group traveling together will often have different survival odds than a first class solo passenger paying a similarly high fare. The interaction between fare, class, and cabin location is where things get interesting. I once spent a long time trying to build a model that treated fare as a continuous variable and kept getting poor performance. The fix was binning fare into quartiles within each class rather than across the whole dataset. That single change reduced overfitting and made the coefficients much more interpretable. Anyone who skips that step will produce results that look clean on paper but don't reflect the actual distribution.
The Embarkation Port Factor
Passengers who boarded at Cherbourg had a higher survival rate than those who boarded at Southampton or Queenstown. The Cherbourg passengers were predominantly first class, which confounds the interpretation. When you control for class, the port effect shrinks considerably but doesn't disappear entirely. There may have been slight differences in boarding speed or route familiarity, but the signal is weak enough that you should treat it as a minor adjustment rather than a driving factor. About 20% of the full passenger manifest has missing age values. Cabin data is missing for nearly 80% of records. This isn't a minor cleanup issue. If you drop rows with missing cabin values, you're left with a heavily biased sample that overrepresents first class passengers and skews every survival estimate upward. The practical workaround I use is multiple imputation for age based on class, sex, and fare quartile. For cabin, I treat it as a categorical feature with a dedicated missing category rather than imputing values. This preserves the information content of the missingness itself, which turns out to be meaningful. People who didn't have assigned cabins were mostly crew or very poor third class passengers, and both groups had lower survival rates.
What The Data Gets Wrong
The biggest limitation of any datadriven Titanic analysis is that it reduces human decisions to statistical patterns. The dataset can tell you that a third class male infant had a higher conditional survival probability than a third class adult male. It cannot tell you why. The reasons involve individual moments of panic, officer discretion, physical strength, and sheer luck on a dark freezing night. Another blind spot is the crew. The official crew manifest counts vary between sources. Some lists include 885 crew members while others reach 900. The survival calculations shift slightly depending on which count you use, and the discrepancy matters more when you're publishing something that other researchers will cite.
Bottom Line On The Numbers
549 survived out of roughly 2,224 people on board. Women survived at about three to four times the rate of men. First class survived at roughly twice the rate of third class. Children under five had a meaningful survival advantage. Family size had a moderate positive effect up to a point. Missing data is substantial and handling it poorly will distort your results. The numbers are useful for understanding broad patterns, but they cannot capture the chaotic reality of what actually happened that night.
