📚 Homeschooling Data: Release Cycles and Statistical Definitions | 在家教育数据的发布周期与统计口径
Homeschooling is one of the few education populations that no state counts directly. Every published number is a by-product of an administrative process, released on a timetable that serves the administration rather than the analyst. Understanding release cycles and statistical definitions is therefore the difference between reading a figure correctly and misreading it completely.
在家教育是少数几个没有任何政府直接完整统计的教育群体之一。所有已公布的数字都是行政流程的副产品，其发布节奏服务于行政需要，而不是研究需要。因此，理解发布周期与统计口径，是"正确解读一个数字"和"彻底误读一个数字"之间的分界线。
1. Why Homeschooling Data Is Hard to Count | 为什么在家教育数据难以统计
Most education statistics are built from school rolls. Homeschooling is defined by the absence of a school roll, so it sits at the edge of every administrative system rather than at the centre of any single one. There is no natural register that must record these children.
大多数教育统计数据来自学校学籍名册。而在家教育恰恰以"没有学籍"为界定特征，因此它位于每个行政系统的边缘，而不在任何单一系统的中心。没有任何一个天然的登记册必然记录这些学生。
Registration is voluntary in many jurisdictions and enforcement is limited, so the true population is unknown. Every published figure should be treated as a lower bound on an unobservable total, not as a complete census.
在许多地区，登记是自愿的，执法力度也有限，因此真实总体规模未知。所有已发布数字都应被视为某个无法观测总量的下限，而不是一次完整的普查。
Children move between school and home during a year, so a snapshot count and a full-year count answer different questions.
学生在一年之内会在学校与家庭之间往返流动，因此"单日快照"与"全年累计"回答的是完全不同的问题。
Administrative data are collected to run a system, not to measure it, so definitions change when policy changes.
行政数据是为了运行一个系统而收集的，不是为测量它而收集的，因此政策一变，口径就变。
Small counts are often suppressed for confidentiality, which removes exactly the cells analysts most want.
小样本数值常因保密要求被抑制，而这恰恰删掉了分析者最想要的那些单元格。
2. Defining the Scope: What Counts as Homeschooling | 口径界定：什么算作在家教育
Before comparing any two figures, identify the definition behind each. Elective home education, education otherwise than at school, remote schooling and flexi-schooling are four different populations that are frequently added together by mistake.
在比较任何两个数字之前，先确认各自背后的定义。选择性在家教育、校外替代性教育、远程学校和弹性上学是四个不同总体，却常被错误地加总在一起。
The decisive test in most systems is whether the child remains on a school roll. If the child is enrolled in a school, a distance-learning provider or an online academy, the child is normally counted within that institution, not within the homeschooling population.
在多数制度中，决定性的判断标准是：该学生是否仍在学校学籍名册上。如果学生已注册于某所学校、某远程教育机构或某在线学校，通常会被计入该机构，而不计入在家教育群体。
| Term | Typical definition | Counted as homeschooling? |
| Elective home education (EHE) | Parent chooses to educate at home; child not on a school roll | Yes |
| Education otherwise than at school (EOTAS) | Authority arranges off-site provision for medical or behavioural reasons | Usually no, reported separately |
| Remote or online schooling | Child enrolled with a provider and still on a roll | No |
| Flexi-schooling | Part-time school plus part-time home education | Partly; depends on the rule applied |
A change of definition can produce a jump in the series that looks like a behavioural change but is purely a measurement change. Always check the methodology note before interpreting a trend.
口径的一次变化就可能造成序列的跳变，看起来像是行为变化，实际上纯粹是测量变化。在解读趋势之前，务必查阅方法说明。
3. Counting Units: Children, Families and Households | 计数单位：学生、家庭与住户
The unit of count must be stated explicitly. A count of children, a count of families and a count of households can differ by a factor of two or more, because homeschooling often clusters within families.
计数单位必须明确说明。学生数、家庭数与住户数可能相差两倍甚至更多，因为在家教育往往在家庭内部聚集出现。
Mean children per homeschooling family: x̄ = Σxᵢ ÷ n
If the mean is 1.8, then 10,000 families correspond to roughly 18,000 children. Reporting the family count as though it were a child count understates the population by about 44%.
如果平均值是 1.8，那么 10,000 个家庭大约对应 18,000 名学生。若把家庭数当作学生数发布，会使总体被低估约 44%。
Household-based denominators are needed for household-level rates, and child-based denominators for child-level rates. Never mix them.
家庭层面的比率需要以住户为分母，学生层面的比率需要以学生为分母，二者绝不可混用。
A snapshot count gives prevalence on one day; a cumulative count gives incidence over a period. They are not interchangeable.
快照计数给出的是某一天的时点流行度，累计计数给出的是一段时期内的发生量，二者不可互换。
4. Reference Periods and Census Dates | 参考期与统计基准日
Official school censuses typically use a single reference day, such as the third Thursday in January or a specified day in October. Only children whose status is correct on that exact day are counted, which makes the series sensitive to timing.
官方学校普查通常采用单一统计基准日，例如一月的第三个星期四或十月某指定日期。只有在该日状态正确的学生才被计入，因此该序列对时间点非常敏感。
The three common reference periods answer three different questions and produce three different numbers:
三种常见参考期回答三个不同问题，并产生三个不同数字：
Point-in-time: how many children were educated at home on the census day?
时点口径：普查基准日当天有多少学生在家接受教育？
Period: how many were educated at home at any point during the academic year?
时期口径：本学年内任意时点有多少学生曾在家接受教育？
Person-time: how many child-years of home education were delivered?
人时口径：总共提供了多少"学生年"的在家教育？
Person-years = Σ (number of children × duration in years)
Because churn is high, the period figure is usually much larger than the point-in-time figure. Quoting one against the other is a common but serious error.
由于流动性很高，时期口径的数值通常远大于时点口径。把两者相互比较是常见但严重的错误。
5. The Release Cycle: From Collection to Final Data | 发布周期：从数据采集到终值
Every statistical release follows the same chain: collection, validation, cleaning, aggregation, quality assurance, publication, then revision. Each stage adds weeks or months of lag between the event and the published number.
每一次统计发布都遵循同样的链条：采集、核验、清洗、汇总、质量保证、发布，然后修订。每个环节都在事件发生与数字公布之间增加数周或数月的时滞。
School censuses are commonly collected termly, while headline homeschooling analyses are published annually. This mismatch matters: the annual release may reflect a return submitted five to nine months earlier.
学校普查通常按学期采集，而关于在家教育的核心分析通常按年发布。这种错配很关键：年度发布所依据的数据，可能是五到九个月前提交的申报表。
| Stage | Typical lag | Data status |
| Collection closes | Reference day + 4 to 8 weeks | Raw returns |
| Validation and queries | + 4 to 12 weeks | Provisional |
| First statistical release | + 3 to 9 months | Provisional, flagged |
| Revised and final series | + 12 to 24 months | Final or near-final |
Good practice is to publish a pre-announced calendar so that users know when data will appear, and to label every figure with its status.
良好做法是提前公布发布日历，让使用者知道数据何时出现，并为每个数字标注其状态。
6. Revision Policy and Data Vintages | 修订政策与数据版本
A vintage is the version of a dataset at a particular release date. The same reference year can carry different values across vintages, because late returns arrive, duplicates are removed and errors are corrected.
数据版本是指某次发布日期时该数据集的具体形态。同一个参考年份在不同版本中可能有不同数值，因为迟到的申报会补入、重复记录会被剔除、错误会被更正。
A simple revision measure compares the first published value with the latest value:
一个简单的修订度量是把首次发布值与最新值进行比较：
Revision = (Vlatest − Vfirst) ÷ Vfirst × 100%
A revision of plus 8% means the first estimate understated the final count by eight percentage points. If revisions are systematically positive, the release cycle is biased downward, not merely noisy.
若修订值为正 8%，说明初值比终值低估了八个百分点。如果修订长期系统性为正，那么发布周期带来的就不只是噪声，而是向下的偏差。
Always state the vintage when quoting a number, especially in a time series.
引用数字时务必注明数据版本，尤其是在时间序列中。
Never compute a growth rate by dividing a provisional figure by a revised figure.
绝不要用初值除以修订值来计算增长率。
7. Timeliness versus Accuracy | 及时性与准确性的权衡
Statistical agencies face a genuine trade-off. Publishing early gives policymakers timely information but risks larger revisions; publishing late gives more accurate data but may arrive after the decision has already been made.
统计机构面临真实的权衡：早发布能为决策者提供及时信息，但修订幅度可能更大；晚发布数据更准确，却可能在决策完成之后才到位。
Speed is often bought with three concessions: smaller samples, more imputation for missing returns, and looser validation rules. Each concession widens the confidence interval around the published figure.
提速通常以三项让步为代价：样本更小、对缺失申报的插补更多、核验规则更宽松。每一项让步都会扩大已发布数字的置信区间。
For users, the practical rule is simple: use provisional data for direction and final data for magnitude, and never mix the two within one chart.
对使用者而言，实用规则很简单：初值用来看方向，终值用来看规模，且绝不在同一张图中混用两者。
8. Response Rates, Coverage and Non-Response Bias | 应答率、覆盖率与无应答偏差
Two rates are routinely confused. The response rate measures how many units replied; the coverage ratio measures how much of the true population the data actually capture. A survey can have a high response rate and still cover only a fraction of the population.
有两个比率常被混淆。应答率衡量有多少单位作出了回复；覆盖率衡量数据实际捕捉到了真实总体的多大比例。一项调查可能应答率很高，却仍只覆盖总体的一小部分。
Response rate = (valid responses ÷ eligible units) × 100%
Coverage ratio = registered count ÷ estimated true population × 100%
Non-response bias depends on both the size of the non-responding group and how different it is from respondents:
无应答偏差既取决于未应答群体的规模，也取决于它与应答者的差异程度：
Bias ≈ (ȳ_non-response − ȳ_response) × (nnon-response ÷ ntotal)
Families who do not register are precisely those most likely to be missing from administrative data, and they may differ systematically in socioeconomic profile. This makes the bias directional rather than random.
未登记的家庭恰恰是最可能从行政数据中缺失的群体，且其社会经济特征可能存在系统性差异。这使偏差具有方向性，而不是随机噪声。
9. Rates, Denominators and Small-Count Instability | 比率、分母与小样本波动
A count alone is rarely informative. Rates require a denominator that matches the numerator in definition, geography and time.
单纯的计数很少具有信息量。比率要求分母在口径、地理范围和时间上都与分子相匹配。
Rate per 1,000 pupils = (homeschooled pupils ÷ total pupils) × 1,000
Percentage change = (Vₜ − Vₜ₋₁) ÷ Vₜ₋₁ × 100%
For small populations, year-on-year changes are dominated by random variation. A rise from 12 to 15 cases is a 25% increase but may be statistical noise rather than a genuine trend.
对于小规模总体，同比变化主要由随机波动主导。从 12 例升至 15 例，虽是 25% 的增长，却可能是统计噪声而非真实趋势。
SE(p̂) = √(p̂(1 − p̂) ÷ n)
95% confidence interval = p̂ ± 1.96 × SE(p̂)
Where comparisons involve authorities of very different sizes, use rates and confidence intervals together, and avoid ranking small authorities on point estimates alone.
当比较涉及规模差异极大的地区时，应同时使用比率与置信区间，避免仅凭点估计对小地区进行排名。
10. Comparability Across Jurisdictions and Exam Boards | 跨地区与跨考试局的可比性
Comparability fails on at least four dimensions: legal definition, reference period, academic year boundaries and counting unit. Two systems can both publish a homeschooling total with almost no overlap in meaning.
可比性至少在四个维度上失效：法律定义、参考期、学年边界与计数单位。两个体系可能都发布了在家教育总数，其含义却几乎没有交集。
Academic years differ: September to August, January to December, and August to June are all in use.
学年边界不同：九月至八月、一月至十二月、八月至六月都被使用。
Compulsory school age ranges differ, changing the eligible population and therefore the denominator.
义务教育年龄
Published by TutorHao | Exam Prep Revision Series | aleveler.com
广告 / Sponsored英文原版 Oxford IB Diploma Programme IB Prepared Mathematics An…券减¥45.0¥296原价 ¥388券后价 · 以淘宝页面为准数学 备考教辅 · 正版现货 · 与本文同科同级别查看详情 / View on Taobao广告 / Sponsored培生爱德思考试局 Alevel 考试系列 Pearson Edexcel International A Level …券减¥5.0¥59.2原价 ¥268券后价 · 以淘宝页面为准数学 备考教辅 · 正版现货 · 与本文同科同级别查看详情 / View on Taobao更多咨询请联系16621398022（同微信）
备考资讯每日更新，关注不错过考纲变化 · 科目分析 · 学习方法
在微信里长按二维码识别关注
或截图后用「扫一扫 → 相册」选择本图公众号：tutorhao
