정량 · 완주

작은 설문 하나를 끝까지: 빈도에서 회귀까지

지어낸 설문 60명 · 여섯 마디 — 빈도 → 기술통계 → t검정 → 일원 ANOVA → 상관 → 회귀

연구 질문 — 점수는 성별·부서에 따라 다른가, 그리고 무엇이 점수를 예측하는가?

이 사례는 영어입니다 — 원자료·표·APA 문장이 영어로 되어 있습니다.

원자료 walkthrough_survey.csv · 60행

지어낸 자료 — scripts/walkthrough/make_sample.py 가 만든 60행 설문 샘플 자료(안내 글 세 편과 따라 하기가 쓰는 그 샘플 자료). 표와 APA 문장은 backend/tests/test_en_sample_walkthrough_en2.py 가 엔진을 영어로 실제 돌려 낸 실측치입니다.

idgenderagedepartmentsatisfactionscore
R001F28Sales44.12
R002M20Sales32.93
R003F33Support21.91
R004M32Sales32.46
R005F46Support54.9

완주 여정 — 단계별로 실제로 만든 것

  1. ① 빈도 — 누가 답했나

    먼저 셉니다. 성별과 부서에 각각 몇 명이 있는지, 빠진 칸은 없는지. 이 마디를 건너뛰면 뒤의 모든 검정이 «몇 명 위에서 돌았는지» 모른 채 나옵니다.

    A frequency analysis was conducted on gender and department (N = 60). For every variable, valid n = 60 and missing = 0. For gender (2 categories), "F, M" each appeared equally often (30, 50.0% of valid cases). For department (3 categories), the most frequent category was "Research" (24, 40.0% of valid cases). No missing data were present.
  2. ② 기술통계 — 점수의 생김새

    점수 한 열의 평균·표준편차·중앙값·사분위·최솟값·최댓값과 왜도·첨도를 봅니다. 분포가 한쪽으로 몰려 있으면 뒤에서 고를 검정이 달라집니다.

    Descriptive statistics were computed for score (N = 60). For score, M = 3.75, SD = 0.77, Mdn = 3.96, range = 1.73–5.00. Skewness = -0.57 and kurtosis = -0.05, within normal range. No missing data were present.
  3. ③ 독립표본 t검정 — 성별에 따라 다른가

    두 집단의 평균을 비교합니다. 등분산 가정은 Levene 검정이 함께 재고, 결과에는 효과크기(Cohen’s d)와 평균차의 신뢰구간이 붙습니다.

    Independent-samples t-tests compared score across levels of gender (N = 60). For score, F (M = 4.11, SD = 0.66) and M (M = 3.39, SD = 0.71) differed significantly, with a higher mean in F, Student's t(58) = 4.06, p < .001, Cohen's d = 1.05, mean difference = 0.72, 95% CI [0.36, 1.07]. This was a large effect by Cohen's (1988) criteria. Levene's test indicated equal variances.
  4. ④ 일원 ANOVA — 부서에 따라 다른가

    집단이 셋이면 t검정을 세 번 하지 않고 한 번에 봅니다. 효과가 유의할 때 어느 쌍이 갈라지는지는 사후검정(Tukey HSD)이 짚어 주고, 그 표도 아래에 함께 실립니다.

    One-way ANOVAs examined score across levels of department (N = 60). For score, the difference was not statistically significant across groups, F(2, 57) = 1.77, p = .180, η² = .06. Levene's test indicated equal variances (p = .886). Group statistics: Research (M = 3.96, SD = 0.71), Sales (M = 3.51, SD = 0.80), Support (M = 3.70, SD = 0.79).
  5. ⑤ 상관 — 무엇이 함께 움직이나

    변수 셋을 한 번에 비교하면 쌍이 셋입니다. 그래서 문장이 아니라 상관행렬로 냅니다 — 칸마다 r과 p가 함께 있습니다.

    A Pearson correlation analysis was conducted (N = 60). score and age showed a significant positive relationship, r = .26, 95% CI [.01, .49], p = .041. This was a small effect by Cohen's (1988) criteria. score and satisfaction showed a significant positive relationship, r = .87, 95% CI [.79, .92], p < .001. This was a large effect by Cohen's (1988) criteria. For the other 1 pair, the relationship was not statistically significant.
  6. ⑥ 회귀 — 무엇이 점수를 예측하나

    나이와 만족도로 점수를 예측합니다. 성별은 글자라 엔진이 회귀에서 빼는데, **빼면서 그렇게 말합니다** — 조용히 지우지 않습니다(아래 각주).

    A multiple linear regression was conducted to predict score from age, satisfaction. The overall model was significant (R² = .77, adjusted R² = .76, F(2, 57) = 93.64, p < .001). This was a large effect by Cohen's (1988) criteria. The predictors accounted for 76.7% of the variance in score. For age, the effect was not statistically significant (β = .11, p = .083). satisfaction significantly predicted score (positively; B = 0.72, SE = 0.06, β = .85, t(57) = 13.05, p < .001). Multicollinearity was assessed using the variance inflation factor; the maximum VIF was 1.03 (age), below the threshold of 10, so multicollinearity was not a problem. · Excluded variables: gender — not numeric, so they were left out.

핵심 산출

① 빈도 — Frequency table

Total N = 60, Total missing = 0 · Cumulative percent: based on valid responses

VariableCategoryFrequencyPercentValid PercentCumulative Percent
genderF3050.050.050.0
genderM3050.050.0100.0
departmentResearch2440.040.040.0
departmentSales1728.328.368.3
departmentSupport1931.731.7100.0
Total60100.0100.0100.0

② 기술통계 — Descriptive statistics

Total N = 60, Total missing = 0

VariableNMeanStd. DeviationVarianceStd. ErrorMedianQ1Q3MinimumMaximumSkewnessKurtosis
score603.750.770.590.0993.963.234.271.735.0-0.567-0.052

③ 독립표본 t검정 — Mean comparison by gender

score: Levene p=.399 · Effect sizes were interpreted as small, medium or large following Cohen's (1988) criteria.

VariableGroupNMeanSDtdfp
scoreF304.110.6574.05858< .001
M303.390.713

④ 일원 ANOVA — score — Post hoc comparisons (Tukey HSD)

ComparisonMean differenceStd. Errorp95% CIDecision
Research vs Sales0.4450.170.163[-0.134, 1.025]No clear difference
Research vs Support0.2560.165.520[-0.305, 0.818]No clear difference
Sales vs Support-0.1890.179.737[-0.800, 0.421]No clear difference

④ 일원 ANOVA — Mean comparison by department

score: Post hoc: no pair showed a difference

VariableGroupNMeanSDSSdfMSFp
scoreResearch243.960.7072.04121.0211.767.180
Sales173.510.796
Support193.700.792
Within groups32.921570.578
Total34.96359

⑤ 상관 — Pearson correlation matrix

N = 60 (pairwise) · Each cell: r (p) · Effect sizes were interpreted as small, medium or large following Cohen's (1988) criteria.

Variableagesatisfactionscore
age1
satisfaction0.176 (.178)1
score0.264 (.041)0.868 (< .001)1

⑥ 회귀 — Factors predicting score

Dependent variable: score · N = 60 · Model summary rows: R² and adjusted R² sit in the value column; the model F sits in the test-statistic and significance columns. · Effect sizes were interpreted as small, medium or large following Cohen's (1988) criteria.

VariableBSEβtp
(Constant)0.7740.2403.227.002
age0.0070.0040.1151.765.083
satisfaction0.7180.0550.84813.047< .001
R²0.767
Adjusted R²0.758
Model F(2, 57)93.643< .001

대표 인용

모든 인용은 원자료에 실재하는 문장입니다 (발췌 실재 검증 통과).

    논문형 표

    복사하면 편집 가능한 사본이 작업대에 열립니다 — 이 전시는 그대로 남습니다.

    A small survey, start to finish: from frequencies to regression | Statory