## Distribution summary for the numeric columns Nine columns are analytically numeric. Seven are stored as text (`VARCHAR`) and had to be parsed, and `Productivity_Score`, `Accuracy_Rating` and `Satisfaction_Score` are already numeric. The summary table above (the final profiling query) shows min, quartiles, mean, p95, max and standard deviation for each. The figures below come from that result. | Column | Valid n | Min | Median | Mean | Max | Std dev | |---|---|---|---|---|---|---| | Age | 263 | 16 | 27 | 30.18 | 97 | 12.47 | | Monthly_Income | 290 | 0 | 521.5 | 2,304.38 | 250,000 | 14,743.44 | | AI_Usage_Hours_Per_Day | 271 | 0.5 | 2.2 | 3.01 | 23 | 2.45 | | Monthly_AI_Cost | 288 | 0 | 0 | 9.84 | 100 | 13.75 | | Time_Saved_Hours_Per_Week | 272 | 0 | 5.05 | 5.81 | 19.4 | 4.11 | | Work_or_Study_Hours_Per_Day | 279 | 2.1 | 7.2 | 7.29 | 20 | 2.14 | | Productivity_Score | 284 | 1.6 | 5.0 | 5.23 | 10 | 1.82 | | Accuracy_Rating | 285 | 2 | 4 | 3.43 | 5 | 0.86 | | Satisfaction_Score | 293 | 3 | 6 | 6.15 | 10 | 1.52 | ### What the spreads show - **Monthly_Income is extremely skewed.** The mean (about 2,304) is 4.4 times the median (521.5). The 95th percentile is 6,399, but the maximum is 250,000, and that outlier drives the 14,743 standard deviation. Use the median or percentiles here, not the mean. - **Monthly_AI_Cost is mostly zero.** The median and 25th percentile are both 0, and the 75th percentile is 20. The mean of 9.84 comes from a minority of paying users, with a 95th percentile of 30 and a maximum of 100. - **AI_Usage_Hours_Per_Day is right-skewed.** The median is 2.2 hours and the mean 3.01. The interquartile range is 1.4 to 4.1, and the 95th percentile is 7.55. The 23-hour maximum is implausible for a daily figure and looks like a data-entry error. - **Age is right-skewed with a suspicious maximum.** The middle half is 22 to 35 and the median is 27, while the maximum is 97, which is a likely outlier. - **Work_or_Study_Hours_Per_Day is symmetric.** The mean (7.29) and median (7.2) are nearly equal, with a middle half of 6.1 to 8.4. Only the 20-hour maximum is an outlier. - **Time_Saved_Hours_Per_Week is mildly right-skewed.** The median is 5.05 and the mean 5.81. The middle half is 2.7 to 7.75, and values reach 19.4. - **The three rating scales are tightly clustered and roughly symmetric.** - Productivity (1.6–10) has a median of 5.0 and a mean of 5.23. - Satisfaction (3–10) has a median of 6 and a mean of 6.15. - Accuracy (2–5) has a median of 4 and a mean of 3.43. Its small standard deviation (0.86) shows it is a compressed 1–5 style scale, with most values at 3 or 4. ### Data quality caveats - **Missing or unparseable values:** these range from 7 (Satisfaction) to 37 (Age) out of 300 rows. The other counts are AI usage 29, Time saved 28, Work/study hours 21, Productivity 16, Accuracy 15, AI cost 12 and Income 10. - **Parsing sensitivity:** an alternative parse treated more text values in two columns as missing. That gave 273 valid values for `Monthly_AI_Cost` (mean 9.76) and 284 for `Monthly_Income` (mean 2,328.96, median 520). The differences are small and the conclusions do not change. I used the fuller parse above. - **Privacy:** `Monthly_Income` is flagged as person-identifying, so only aggregates are reported here.