Making More Sense of School Data 分析學校自評數據 Dr. Soh Kay Cheng 蘇啟禎博士 Consultant (QA) EMB, Hong Kong SAR jensoh@singnet.com.sg Soh KC(2006) Making sense of school data 1 Why do we need data (statistics)? 為何要用數據(統計數字)﹖ Is a special language Ensures objectivity Avoids misunderstanding Facilitates communication 它是特殊的語言 它確保客觀 它避免誤解 它促進溝通 Soh KC(2006) Making sense of school data 2 Data for School Self-evaluation 學校自評的數據 Describing status 描述現狀 例如 KPM 4教師的學歷與經驗 e.g., KPM4 Teacher qualification & experience 描述見解 Describing views 例如 KPM 2教師對領導的評價 e.g., KPM 2 Staff’s views on leadership 描述表現 Describing performance 例如KPM 22學生出席率 e.g,, KPM22 Student attendance 描述趨勢 Describing trends 例如KPM 9過去三年學生閱讀習 e.g., KPM9 Students’ reading habits of past 3 years 比較技術 Techniques for comparisons 例如 KPM 4教師學歷與本港參考 數據比較 e.g., Compare KPM4 Teacher qualification with Hong Kong’s Reference Data Soh KC(2006) Making sense of school data 3 Describing needs reference data 描述需要參考數據 孤立的數據(如%) A statistic (e.g., %) standing 沒有多大意義甚至毫 alone has little or no meaning. 無意義。數據要有意 One or more points are 義﹐必須另有數據作 needed for meaningful 為參考。 interpretation of the given 參考數據可能是隱含 statistic. 的﹐假設數據使用者 Reference points may be 已有共識. implicit, assuming the users have common understanding. 隱含的數據必須明朗 化﹐明確指出參考數 Implicit reference points need 據的性質。 be made explicit. Soh KC(2006) Making sense of school data 4 Good for a laugh A professor of statistics meets a colleague on his way to the lecture. The colleague heartily greets him, “Good morning, professor!” and then respectfully asks, “How is your wife?” And the professor absent-mindedly says, “….?” Soh KC(2006) Making sense of school data 5 Describing status 描述現狀 Soh KC(2006) Making sense of school data 6 How popular are these ECAs? 這些課外活動受歡迎嗎 課外活動學會/小組的數目 學會 / 小組 組數 課外活動學會/小組的數目 學會/ 小組 組數 人數 每組平均人數 學術 1 學術 1 50 50 體育 9 體育 9 270 30 藝術 5 藝術 5 50 10 興趣 9 興趣 9 90 10 社會服務 12 社會服務 12 240 20 The number of groups may not reflect the popularity of each activity. The nature of the activities may have an impact on their popularity. In addition to the number of units, it is good to report also the group sizes to reflect their populatiry. 每種活動的組別數目不一定反映活動受歡迎的程度。 活動是否受歡迎﹐和活動的性質有關。 除報告組數外﹐同時也報告各組人數﹐更能反映活動受歡迎的程度。 Soh KC(2006) Making sense of school data 7 Box-and-whisker plot as reference 盒鬚圖作為參照 EMB issued Reports on Key Performance Measures Reference Data 2003/2004 for both primary and secondary schools. The median is the mid-point of a distribution. 50% of the schools are at or above (below) the median. Top 25% are at or above the 75th percentile. Likewise, bottom 25% are below the 25th percentile. The middle 50% are between the 25th and 75% percentiles. 教統局印發2003/2004年度 中小學校表現評量參考數據 報告。 中位數將學校分為上下兩個 群組﹐各有50%。 最高的25% 在75百分位數或 以上。最低的25%在25百分 位數之下。中間的50%在兩 者之間。 Soh KC(2006) Making sense of school data 8 Use of box-and-whisker plot 盒鬚圖的運用 百份率 Percentage (%) The percentage of subject-trained Chinese and Mathematics teachers of school are both 85%. How would you report on these? For Chinese, the school is at the median of reference data For Mathematics, the school is among the top quarter. 本校已接受專科訓練的中文和數學教師 的百分率均為 85%。這情況怎麼報告? 已接受專科訓練的三個核心科目教師的百分比 % of Subject-trained Teachers 100 90 * 85% * 80 70 60 中文教師方面,本校位列於參考數據的 中位數。 50 數學教師方面,本校位列於最高的四分 之一之內。 40 中文科教師 Subject-trained Chi Teachers Soh KC(2006) Making sense of school data 數學科教師 Subject-trained Maths Teachers 9 Description is not evaluation 描述不等於評估 KPM3 教師專業發展 教師參與持續專業發展的平均時數 82.2 校長參與持續專業發展的時數 207.0 教學人員(包括教師與校長)在持續專業發展方面的平均預算支出 $127.3 教學人員(包括教師與校長)在持續專業發展方面的平均實際支出 $96.7 Is it good that the teachers spent 82.2 hours and HK$127.3 on professional development? What about the principal? Reference points needed could be Hong Kong norm, school’s past records, or pre-determined targets. 教師花費82.2小時和港幣127.3元於專業發展,理想嗎?校長方面呢? 其他可用的參考資料,如香港常模、本校往年數據、或預定目標。 Soh KC(2006) Making sense of school data 10 Good to have another laugh A young man boasts about his wife and says, “My wife has a perfect figure of 100, it’s 38-24-38.” His middle-age friend, not wanting to lose face, says calmly, “My wife has a perfect figure adding up to 100, too. It is ….” Soh KC(2006) Making sense of school data 11 Describing views 描述見解 Soh KC(2006) Making sense of school data 12 Scale and dispersion 尺度與變異 KPM02 教職員對學校領導層的觀感 校長的領導能力與態度 4.01 副校長的領導能力與態度 4.07 中層領導人員的領導能力與態度 4.01 Five-point scale was used. This fact is implicit and needs be made explicit. Should it be a 10point scale, the interpretation would be rather different. The two means of 4.01 give the impression that the Principal and the Middle Management were equally favorably evaluated. What if the SD for Principal is 0.06 and that for the Middle Management 1.12? 報告採用五度量表。尺度應 該說明。假如所用的是十度 量表,則各階層領導所得的 評估便非常不同。 兩個平均數4.01似乎顯示校 長與中管理層得到同樣的好 評。假如校長的標準差(SD) 是0.06,而中管理層的是1.12, 應作何解釋? Soh KC(2006) Making sense of school data 13 Poverty is relative A rich American wanted to show his son how rich they were. He took the little boy to a seaside village in the South. When everything was over, the father asked, “Son, what have you learned from this trip?” The boy said, “Oh, yes! We keep one dog and they have four. Sitting at our patio, the view ends at the gate 50 yards away, but at their, there is no end to the horizon. We built walls to protect ourselves, they have friends to protect them….Thank you father, for showing me how….” Soh KC(2006) Making sense of school data 14 Describing performance 描述表現 Soh KC(2006) Making sense of school data 15 Absolute and relative standards 絕對標準與相對標準 學生從學校圖書館借用資料頻數(中一至中三) 1 每周一次或以上 24.2% 2 每兩周一次 22.1% 3 每月一次 36.5% 4 每月少於一次 15.8% 5 從不 1.4% • • Relative standard: 36.5% is the mode. 相對標準 • • Absolute standard: 46.3% borrowed at least once in two weeks. 絕對標準 預期目標:每兩週借 用資料至少一次的學 生有50%或以上。 實際上﹐每兩週借用 資料至少一次的學生 有46.3%﹐接近預期 目標50%。 。 借用層次有五個。其中最高的是“每月一次”。 因此﹐學生借用資料頻數的眾數(36.5%)為“每 月一次”。 Soh KC(2006) Making sense of school data 16 The dangerous average Mr. Dumb Bell jumped into the sea from a jetty and got a big hump on his forehead, because the sign board says: First 30 meters, average depth 5 meters! Soh KC(2006) Making sense of school data 17 Describing trends 描述趨勢 Soh KC(2006) Making sense of school data 18 Evaluation varies with reference data 評價隨參考數據而變 校內學科測驗成績 學科 2003 2004 2005 中文 41.2 55.0 52.3 英文 22.8 44.7 35.3 數學 13.1 33.9 39.3 Relatively speaking, there was an improvement in 2004 over 2003 in all three subjects, but, there is a retrogression from 2004 to 2005 in the two languages. Assumption: Papers of the three years are equivalent. Need HK norms for the three years for proper interpretation. 60 50 40 30 20 10 0 2003 2004 中文 英文 2005 數學 相對而言,2004的成績比2003的好, 有進步。但是,從2004 到2005, 兩 語文科有退步的現象。 假設:三年的考卷難度相同。 正確的解釋需要三年的香港常模作 為參考資料。 Soh KC(2006) Making sense of school data 19 Effect Size 效果强度 學校數據 參考數據 比參考數據 課程策劃與組織 3.65 3.54 +0.11* 課程管理 3.70 3.57 +0.21* 教學策略和技巧 3.65 3.45 +0.20* *比參考數據 +/-0.1或以上 Comparison with Reference Data is a good effort. However, +/-0.1 將本校的情況和參考數據比較, 是好的做法。 但是﹐以+/-0.1 is arbitrary. 為臨界值﹐似乎武斷。 If SDs are available, then effect sizes can be calculated for more 如有標準差﹐可轉化為效果強 meaningful interpretation. 度(Effect size),更有意義。 Soh KC(2006) Making sense of school data 20 Conversion to effect size 效果強度的轉化 School Mean Reference Data Mean (SD) Effect Size 課程策劃與組織 3.65 3.54 (0.25) 0.44 課程管理 3.70 3.57 (0.18) 0.72 教學策略和技巧 3.65 3.45 (1.15) 0.20 註﹕參考數據是虛擬的。 With reference to the normal distribution, ES can be used to evaluate differences in percentages. ES = (Mean – Norm) / SD 如參照常態分佈﹐百分率可轉化為效果強度﹐以便檢定百分率差異的 意義。 效果強度 = (平均數 – 常模)/ 標準差 Soh KC(2006) Making sense of school data 21 Effect size ‘standards’ 效果強度的‘標準’ Difference in % Effect size Description Below 0.2 Negligible effect 微效果 Between 8 and 18 % Between 0.20 and 0.49 Small effect 小效果 Between 19 and 28% Between 0.50 and 0.79 Between 0.80 and 0.99 1.00 and above Moderate effect 中等效果 Below 8% Between 29 and 34% 35% or more Soh KC(2006) Making sense of school data Large effect 大效果 Very large effect 極大效果 22 Summary 綱要 Use data to ensure objectivity and 1. common frame of mind. 2. 2. Use data to describe status, views, performance, and trends. 3. 3. Use reference data for meaningful 4. interpretation. 5. 4. Use %’s for comparison. 6. 5. Use reference data for evaluation。 6. Data for reference may be the norms, 7. past records, or targets. 7. Two means of the same magnitude may have different meaning. Watch out 8. for difference in variability. 8. Use absolute or relative standard to 9. describe performance 9. Use curves to indicate trends and 10. watch out for tacit assumptions. 10. Use effect size ‘standards’ for objective evaluation of effects. 1. 用數據溝通,力求客觀共識。 用數據描述現狀﹑見解﹑表 現﹑與趨勢。 用參考數據作有意義的詮釋。 用百分比作比較。 用參考數據來評估。 參考的數據可能是常模﹑記 錄﹑或目標。 兩個相同的平均數可能有不同 意義。注意變數的大小。 用絕對標準或相對標準描述表 現。 用曲線圖表達趨勢﹐並注意隱 含的假設。 用效果強度的「標準」進行客 觀的評估。 Soh KC(2006) Making sense of school data 23 Statistics are estimates I asked a statistician for her telephone number and she gave me an estimate. Soh KC(2006) Making sense of school data 24 Comparison Techniques 比較技術 Soh KC(2006) Making sense of school data 25 Purpose School reports always present summary data such as the mean, the SD, etc. Statistical calculators on the Internet can be used to make some of the needed comparisons. This part of the seminar introduces such calculators to enable schools to make finer interpretation. Soh KC(2006) Making sense of school data 26 Comparison with expected performance 與預期表現比較 N 65 Mean 3.71 The result of a survey on KPM2 Teachers’ view on the principal’s leadership. If the expected mean is 3.50, was the principal more favorably evaluated than is expected? This calls for a one-sample t-test and the critical value of p is set at 0.05. SD 0.35 KPM2 教師對校長領導能力的評 估,調查結果如表所示。 如果預期平均數為3.50,校長所 得平均數是否較預期的為高? 這需要用單組t-測加以鑒定,並以 0.05為p的臨界值。 Soh KC(2006) Making sense of school data 27 The result shows that the obtained mean (3.71) is statistically greater than the expected mean (3.50). The principal was evaluated higher than the expected. 統計分析結果顯示,實得的平均數 (3.71)的確高於預期的平均數 (3.50)。校長所得評估的確比預 期的高。 http://glass.ed.asu.edu/stats/analysis/ttest.html Soh KC(2006) Making sense of school data 28 Comparison of two groups 兩群組比較 Teachers N 65 Mean 3.52 SD 0.30 Parents 125 3.75 0.65 The table shows the ratings on KPM11 School culture given by teachers and parents. Did the parents evaluate the school more favorably than did the teachers? 上表顯示教師與家長對KPM11 學校文化的評估。兩者對學校的 確有不同評價嗎? 這必須用獨立t-測加以鑒定,並 以0.05為p的臨界值。 This calls for an independent ttest and the critical p is set at 0.05 Soh KC(2006) Making sense of school data 29 http://glass.ed.asu.edu/stats/analysis/t2test.html The result shows that the difference (0.23) between the two means is unlikely a chance occurrence. Parents did evaluated the school more favourably than did the teachers. 統計結果顯示,兩平均數之差 (0.23) 並非機遇現象。家長對學 校的評價的確高於教師的評價。 Soh KC(2006) Making sense of school data 30 Comparison of many groups 多群組比較 Teachers (N=65) Students (N=120) Parents (N=105) The ratings on KPM 11 School culture as given by teachers, students, and parents are shown above. Did the three groups differ in their rating? This calls for one-way ANOVA (analysis of variance) followed by pair-wise comparisons. Mean 3.51 3.65 3.83 SD 0.35 0.45 0.52 針對 KPM 11 學校文化,教師、 學生、與家長作以上的評估。三 組的評價的確有差異嗎? 這需要用單向變異分析(oneway analysis of variance)鑒 定,並再用配對t-測。 Soh KC(2006) Making sense of school data 31 http://statpages.org/anova1sm.html The analysis shows that p<0.05; there is at least one significant difference among the three means. 分析結果顯示p<0.05,表示至少有 一對的平均數有非機遇的差異。 Soh KC(2006) Making sense of school data 32 http://glass.ed.asu.edu/stats/analysis/t2test.html As there are three pair-wise tests, Bonferroni adjustment is applied to avoid accumulation of error. For three tests, the p-value should be 0.05/3 or 0.0166. To check on these p-values, see the next slide. 有三對平均數作配對比較,必須作 Bonferroni調整,以避免誤差的累 積,而p-值應該是0.05/3 or 0.0166. 要確定這三個p-值,請看 下頁。 Soh KC(2006) Making sense of school data 33 http://department.obg.cuhk.edu.hk/researchsupport/T_Test.asp (65 + 120 - 2) For all three t-values (21.76, 43.82, and 27.84), the corresponding p is 0.0001. The differences among the three means are statistically significance; the differences are very unlikely chance occurrences. 三個 t-數(21.76, 43.82, and 27.84) 配對比較的相應p-值是0.0001. 它們 之間的差異並非機遇現象;其間的確 有差異。 Soh KC(2006) Making sense of school data 34 Comparison with reference data (1) 與參考數據比較(一) Professionally trained Not trained Hong Kong 96% (62) 4% (3) School 90% (59) 10% (7) For KPM 4, the school has 90% of its 65 teachers professionally trained. Is this percentage significantly lower than the Hong Kong Reference Data? This calls for a chi-square test of goodness of fit. The percentages need be converted into frequencies for analysis. 學校的65位教師,有90% 受過專業訓練。這和香港參 考資料比較,有差異嗎? 這需要用卡方測驗來鑒定。 並需先將百分數轉為頻數。 Soh KC(2006) Making sense of school data 35 http://www.georgetown.edu/faculty/ballc/webtools/web_chi.html Soh KC(2006) Making sense of school data 36 Soh KC(2006) Making sense of school data 37 The chi-square test results show p>0.05; the school’s distribution fits the Hong Kong distribution. 卡方測驗結果,p>0.05;顯示學校的 分配情況和香港參考數據相配。 Since the chi-square is not significant, the school is not different from the Hong Kong reference data. Soh KC(2006) Making sense of school data 38 Comparison with reference data (2) 與參考數據比較(二) Master Bachelor Non-degree Hong Kong 23% (15) 66% (43) 11% (7) School 27% (18) 58% (38) 15% (9) For KPM 4, the school reported the percentages of teachers with different qualification as shown above. Does the school’s distribution differ significantly from the Reference Data? 學校教師學歷分佈如表所示。與參 考數據相同嗎? 用卡方測驗鑒定。 This calls for a chi-square test. Soh KC(2006) Making sense of school data 39 Since the chi-square is not significant (p>0.05), the school’s distribution is not different from the Reference Data. 卡方檢查結果,p>0.05 ; 學校的分 佈情況和參考數據的沒有不同。 Soh KC(2006) Making sense of school data 40 Summary 綱要 One-sample t-test for comparing a observed mean to an expected mean. Two-sample t-test for comparing the means of two independent groups. One-way ANOVA for comparing more than two independent means. Significance of t-value for checking the probability of an obtained tvalue. Chi-square test for ascertaining association between membership and performance. 單組t-測:比較實際平均數與 預期平均數。 兩組t-測:比較兩組的平均數. 單向變異分析:比較多過兩組 的差異。 t-值的臨界值:檢查所得t-值 是否機遇現象。 卡方測驗:檢定兩個分佈情況 之間的差異。 Soh KC(2006) Making sense of school data 41 Hyperlinks to calculators One-sample t-test http://glass.ed.asu.edu/stats/analysis/ttest.html Two-sample t-test http://glass.ed.asu.edu/stats/analysis/t2test.html One-way ANOVA http://statpages.org/anova1sm.html Significance of t-value http://department.obg.cuhk.edu.hk/researchsupport/T_Test.asp Chi-square test http://www.georgetown.edu/faculty/ballc/webtools/web_chi.html Soh KC(2006) Making sense of school data 42 He who laughs last, laughs best 最後一笑 Q: As a principal, how do you develop your teachers professionally? A: As a responsible leader, I make sure that everyone of them is busy and works hard. Q: What are they working hard on? A: That does not really matter, as long as they are working non-stop. Q: Could you give me an example? A: For instance, …. Soh KC(2006) Making sense of school data 43 Thank you 謝謝 Soh KC(2006) Making sense of school data 44