Twelve months of service requests: daily, weekly and monthly ask different questions

Raw data civil_complaints_2025-09_2026-08.csv · 12 rows

All five steps shown here are values produced by the statistics engines (a cross-tabulation, an independent-samples t-test, two ANOVAs and a multiple regression). The steps are named “daily” and “weekly”, but the calculations at the heart of the daily and weekly reports — control charts and multiple-comparison correction — are not here, because those engines do not exist yet. What goes out automatically today is the monthly report. / Synthetic demonstration data — not real service requests and not any municipality’s data (the city is written as a placeholder). The twelve-month file comes from a seeded generator in the repository, and every table and sentence was produced by running the repository’s own engines. Those values were then checked cell by cell against an answer sheet computed without the engines, with 0 mismatches; the cell count and every verdict are in the comparison table. Control charts and binomial tests have no engine yet, so those cells are held in an explicit “no engine” layer with the reason written beside them.

The study, step by step — what was actually produced

1. ① Intake structure — does deadline compliance differ by department

   The data is cleaned first: duplicate record numbers removed, blank districts marked unknown, elapsed days recomputed from the dates, and rows whose completion date precedes their intake date pulled out for review. Then department is crossed with deadline compliance — this table is where the three cycles start.

   A chi-square test of independence was performed to examine the association between department and within_deadline (N = 16385). The association was significant, χ²(7, N = 16385) = 934.32, p < .001. Cramér's V = .24. Notable cells: Roads & Transport×N(low), Roads & Transport×Y(high), Culture & Sports×N(low), Culture & Sports×Y(high), Welfare Policy×N(low), Welfare Policy×Y(high).

2. ② Daily — how are things right now

   A single day does not ask whether something changed. It counts what arrived, what is still open, and what is past its deadline. Comparing with yesterday swings wildly by day of week, so the comparison is with the same weekday a week earlier. No hypothesis test runs here — one day is not enough data for one.

   Independent-samples t-tests compared daily_intake across levels of before_after_launch (N = 365). For daily_intake, Before (M = 40.55, SD = 17.50) scored significantly lower than After (M = 54.15, SD = 26.32), Welch's t(319.0) = -5.82, p < .001, d = -0.61, 95% CI [-18.21, -8.99]. Levene's test indicated unequal variances (F = 17.43, p < .001); Welch's correction was applied.

3. ③ Weekly — is something moving

   A week asks whether anything spiked. Compare eight request types against last week and one of them will move sharply by chance alone, so looking at all eight at once has to be accounted for. How large the day-of-week effect is, ANOVA answers with a number.

   One-way ANOVAs examined daily_intake across levels of weekday (N = 365). For daily_intake, there was a significant effect, F(6, 358) = 124.09, p < .001, η² = .68. Levene's test indicated unequal variances (p < .001). Welch's test, which does not assume equal variances, showed a significant effect, F(6, 153.74) = 239.63, p < .001. Post hoc Tukey HSD comparisons showed Fri differed from Mon (p < .001), Fri differed from Sun (p < .001), Fri differed from Sat (p < .001), Thu differed from Mon (p < .001), Thu differed from Sun (p < .001). Games-Howell comparisons showed Fri differed from Mon (p < .001), Fri differed from Sun (p < .001), Fri differed from Sat (p < .001), Thu differed from Mon (p < .001), Thu differed from Sun (p < .001). Group statistics: Fri (M = 56.88, SD = 14.96), Thu (M = 52.25, SD = 13.68), Wed (M = 54.85, SD = 17.31), Mon (M = 71.92, SD = 17.25), Sun (M = 16.44, SD = 4.77), Sat (M = 20.58, SD = 5.03), Tue (M = 58.42, SD = 14.26).

4. ④ Monthly — why is it that way

   Only a month of data supports tests that ask about causes. ANOVA checks whether elapsed days differ by district, and the post-hoc test names which district pair separated. Whether intake volume shifted when app submission opened is also examined here.

   One-way ANOVAs examined elapsed_days across levels of district (N = 16385). For elapsed_days, there was a significant effect, F(3, 16381) = 114.00, p < .001, η² = .02. Levene's test indicated unequal variances (p < .001). Welch's test, which does not assume equal variances, showed a significant effect, F(3, 9030.09) = 96.77, p < .001. Post hoc Tukey HSD comparisons showed East differed from West (p < .001), East differed from New town (p < .001), East differed from Central (p = .014), West differed from New town (p < .001), West differed from Central (p < .001). Games-Howell comparisons showed East differed from West (p < .001), East differed from New town (p < .001), East differed from Central (p = .010), West differed from New town (p < .001), West differed from Central (p < .001). Group statistics: East (M = 4.20, SD = 3.80), West (M = 5.20, SD = 4.53), New town (M = 3.74, SD = 3.30), Central (M = 3.95, SD = 3.63).

5. ⑤ Explanation — the department or the deadline design

   Department and district together explain elapsed days. Then compliance is split again — not by department but by the length of the statutory deadline. Where the same numbers reverse their answer under a different split is the heart of this report.

   A multiple linear regression was conducted to predict elapsed_days from department_Culture & Sports, department_Civil Affairs & Passports, department_Welfare Policy, department_Tax, department_Public Safety, department_Housing Renewal, department_Environmental Management, district_West, district_New town, district_Central. The overall model was significant (R² = .28, adjusted R² = .28, F(10, 16374) = 640.53, p < .001). The predictors accounted for 28.1% of the variance in elapsed_days. department_Culture & Sports was not a significant predictor (β = .00, p = .970). department_Civil Affairs & Passports significantly predicted elapsed_days (B = -1.61, SE = 0.09, β = -.13, t(16374) = -17.00, p < .001). department_Welfare Policy significantly predicted elapsed_days (B = 2.05, SE = 0.09, β = .16, t(16374) = 21.63, p < .001). department_Tax significantly predicted elapsed_days (B = -0.52, SE = 0.12, β = -.03, t(16374) = -4.33, p < .001). department_Public Safety significantly predicted elapsed_days (B = -1.11, SE = 0.10, β = -.08, t(16374) = -11.12, p < .001). department_Housing Renewal significantly predicted elapsed_days (B = 5.30, SE = 0.09, β = .43, t(16374) = 57.68, p < .001). department_Environmental Management significantly predicted elapsed_days (B = -0.72, SE = 0.07, β = -.08, t(16374) = -9.65, p < .001). district_West significantly predicted elapsed_days (B = 0.83, SE = 0.07, β = .09, t(16374) = 11.44, p < .001). district_New town significantly predicted elapsed_days (B = -0.30, SE = 0.07, β = -.03, t(16374) = -4.09, p < .001). district_Central significantly predicted elapsed_days (B = -0.35, SE = 0.07, β = -.04, t(16374) = -4.85, p < .001). Multicollinearity was assessed using the variance inflation factor; the maximum VIF was 1.51 (district_Central), indicating no serious multicollinearity.

Key results

Representative quotes

Every quote is a sentence that exists in the raw data (excerpt-existence check passed).

Paper-ready tables