IB内部评估 数据收集 变量控制 不确定度
IB 科学组的内部评估是一份科学研究报告，学生自己定题、自己做实验、自己写报告，最后由任课教师按统一标准打分。数据收集处在整条链的中间位置，前面连着研究问题，后面连着结论，两头都要靠它撑住。这一环写不实，后面的分析与结论再漂亮也补不回来。IB《Biology guide》在 2023 年 2 月发布，2025 年首次评估，内部评估的框架在物理、化学、生物三科之间保持一致。
The internal assessment in the IB sciences is a single scientific investigation that you design, run and report on your own, marked by your teacher against common criteria. Data collection sits between the research question and the conclusion, and both ends depend on it. Weak data collection cannot be rescued by elegant analysis later.
下面按数据收集的实际顺序展开：先看这项任务在总成绩里的位置，再一步步走到表格、图和不确定度。每一节都对应评分标准里的具体措辞，方便对着改。
内部评估占最终成绩 20% 与四项各 6 分的分法
内部评估占最终成绩 20%，满分 24 分，四项标准各 6 分，分别是研究设计与数据分析、结论、评价。数据收集这件事横跨前两项，做得好等于一次拿到 12 分的空间。IBO 官网给出的时间预算是 10 小时，含要求讲解、实验室取数与教师核对真实性。报告正文上限 3000 词，超出的部分不计分。很多学生以为数据越多越好，其实评分看的是数据与研究问题的匹配度。IB 的指南对三个科学科目使用同一套标准，物理、化学、生物的学生面对的是同一个评分口径。
The IA carries 20% of the final grade and 24 marks, split evenly into research design, data analysis, conclusion and evaluation (IBO Biology guide, first assessment 2025). Data collection falls inside the first two criteria, so a neat lab book with careless processing loses marks twice; the ten hours include teacher checks on authenticity. The 3000-word ceiling is a hard reading limit rather than a target. Plan laboratory time before the deadline, because ten hours disappear quickly once sessions are split across a week.
研究问题要把两个变量写成可测量的量
研究设计这一项要看研究问题是否落在具体情境里，并交代自变量与因变量。可测量的写法是给出量名与单位，比如光照强度对水草氧气释放速率的影响，速率用每分钟气泡数表示。很多学生以为题目写得漂亮就能拿分，评分看的是变量能不能取到数据。因变量要能稳定测三次，读数才会收敛。自变量要能由操作者设定档位，不能靠天气或季节碰运气。IBO 的评分说明还要求给出与研究问题直接相关的背景理论，一句带过不算。两者都定不下来，后面的方法描述再详细也补不回来。
A research question reaches the top band only when both variables are measurable, named with units; vague phrasing such as the effect of light on plants leaves the examiner guessing what was recorded. Pick a dependent variable you can read three times with similar results, so that the spread shows up as noise instead of hiding a trend. If the independent variable cannot be set to fixed levels by hand, the method rests on luck. The context sentence should also name the system being studied, which for a plant investigation means the species and the light source.
初级数据、次级数据与模拟数据的取舍
IBO 指南把数据来源分成动手实验、模拟、数据库、电子表格与野外调查几类。动手实验得到初级数据，数据库拿到的是次级数据，模拟软件如果每次输出都一样就不能用。常见的错误是把数据库整张表搬进报告当自己的数据，这会直接压低研究设计的分数。模拟适合成本高或时间跨度长的题目，前提是输出要有波动。次级数据可以用，只是不能成为主体，主要数据仍要自己动手取。电子表格适合做拟合与误差计算，不适合作数据来源。野外调查要交代样方大小、取样位置与重复。
Primary data from your own hands is the safest route; simulations count as primary only when they return variable output, and a database supplies secondary data that should support rather than replace your measurements. Examiners in practice spot copied tables quickly, because the sampling frame and the rounding never match the stated method. Fieldwork is allowed, provided the site, quadrat size and sampling strategy can be justified in a sentence or two.
自变量的范围、间隔与取值个数
方法一节要说明自变量的范围、间隔与每档测几次。IBO 的评分说明只要求把这些决定讲清楚，并没有写死必须几个取值。多数学校按五档以上布置，间隔要覆盖足够宽，让趋势的两端都看得出来。取值挤在一起时拟合斜率算不准，误差棒还会互相重叠。范围推到仪器量程之外会得到失真的读数，比如光电门在遮光过快时漏计。这一步最容易只写三个档位就开做，回校后补不回来。间隔不等距也可以，只要在方法里说明理由。范围与间隔写清之后，读数的取舍才有依据。
Set a range wide enough to show curvature at both ends, with spacing you can defend in the report; the IBO descriptors ask for the reasoning behind range, interval and frequency rather than a fixed number of levels. Most departments run five or more settings, because three points rarely support a straight-line fit. Readings taken beyond the instrument range look like data but are artefacts. State the interval explicitly in the method, because a reader cannot guess whether five settings were equally spaced.
重复次数、平均值与误差棒
每一档取值要重复读数，通常三次起，取平均并保留离散程度。平均值只缩小随机误差，不修正系统偏差。报告里同时给出标准差或极差，误差棒才有依据。读数若集中在末位上下跳动，说明仪器分辨率就是瓶颈。重复次数要写进方法，评定时才看得到数据是设计出来的。很多学生以为多做几次就能把曲线拉直，其实系统误差不会因为次数增加而消失。取平均前先看一遍原始读数，明显偏离的那一条要回去找原因。
Repeats cut random error but never remove a systematic offset, so a mean taken over many readings can still sit half a division away from the true value. Report the spread next to the mean, and let error bars carry that information into the graph. When every reading differs only in the last digit, resolution rather than method is what limits you.
控制变量要写明怎么控，不只列名字
温度、光照、pH、试剂批号、同一台仪器、同一操作者，这些都要在方法里交代控制手段。写温度控制在室温没有信息量，写用水浴维持在 25 摄氏度正负 1 摄氏度才算交代清楚。控制变量的作用是让因变量的变化只能归到自变量上。漏掉一个关键变量，结论就只能写成相关，不能写成因果。常见的错误是把控制变量列成清单却不写手段，评定时按没有控制处理。同一批试剂与同一支移液管这两条最常被漏。控制写实之后，评价一项里的改进建议才有落点。
Naming a control variable without saying how it was held constant reads as an intention rather than a method; a water bath at 25 degrees Celsius plus or minus one degree is evidence, room temperature is not. Controls decide whether the conclusion can claim causation or only correlation. A single unmanaged factor, such as a fresh batch of reagent, is enough to weaken the comparison. Note who operated the equipment, since a second pair of hands changes readings on any manual timing task.
仪器读数、精度与有效数字
读数精度取决于量程与分辨率，选量程时让读数落在中段附近。数字仪表读到末位，模拟刻度读到最小刻度的一半，这是物理组常用的读数约定。同一列数据的小数位要一致，单位写在表头。有效数字多于仪器能力就是虚假精度，少于仪器能力则丢掉信息。滴定管读到 0.05 立方厘米，报告里写成 23.4 就与仪器不符。这一步最容易在换算单位时把小数位改乱。量筒与移液管的精度差一个数量级，选错玻璃器皿会让整组数据白做。
Pick a range so readings fall near the middle of the scale, then quote values at the resolution the instrument actually offers; a digital display is read to its last digit, an analogue scale to half its smallest division. Keep decimal places consistent down a column and put the unit in the header. More digits than the instrument supports signals invented precision.
不确定度的记录、合成与传递
每条读数都要带不确定度，或至少在数据处理一节交代来源。加减运算里绝对不确定度相加，乘除运算里百分比不确定度相加。多次测量时可以用标准差除以根号 n 估计平均值的标准不确定度。百分比不确定度大的那个量，主导最终结果的精度。用两种方法各算一遍，数量级一致就可以写进报告。很多学生以为不确定度只在结论提一句，数据处理一项的 6 分里它占的份量并不小。仪器允差与随机波动要分开写，混在一起会让误差棒失去意义。
Every reading needs an uncertainty attached, together with a sentence on where the number came from. Absolute uncertainties add under addition, percentage uncertainties add under multiplication, and the largest percentage term dominates the final precision. Estimating the standard uncertainty of a mean as the standard deviation divided by root n keeps the arithmetic defensible. Quote the instrument tolerance separately from the spread of repeats, so that both sources stay visible to the reader.
原始数据表怎么排版才叫清晰精确
数据表要区分原始数据与处理后数据，分作两张表更清楚。列头写清量名、符号与单位，同一物理量的小数位对齐。原始读数不要先四舍五入再录入，处理阶段用副本计算。图形坐标轴要有标签与单位，拟合线说明用的是线性还是曲线拟合。清晰与精确是数据分析评定的两个关键词，缺一项就卡在中档。常见的错误是把平均值写进原始数据表，读者无法判断重复次数。表格编号并在正文里引用，评定时能顺着找到依据。
Two tables work better than one: raw readings stay untouched, while processed values are calculated in a second table. Axis labels with units, a stated fit type and consistent decimal places do most of the work here. Clear and precise are separate descriptors in the IBO criteria, so a tidy layout built on rounded raw data still scores in the middle band.
异常值的判断与舍弃理由
明显的离群点先回实验室找原因，接触不良、气泡、读数看错都很常见。找不到原因时，可以按偏离平均值两倍标准差之类的判据剔除，并在报告里写清判据与剔除数量。把不顺眼的点直接删掉属于数据处理造假，评定时会明显扣分。剔除之后要重新计算平均值与误差。保留原始记录是唯一能自证的办法。这一步最容易因为羞于展示波动而动手改数。异常值出现在低档取值时，往往说明方法本身有缺口，值得回头检查一次。
Investigate outliers before deleting anything, because a loose connection or a trapped bubble usually explains the stray value. If no cause appears, state the rejection rule, such as two standard deviations from the mean, and report how many points went. Silently dropping inconvenient readings is treated as fabrication, and the raw record is what protects you. Keep the flagged readings in an appendix and explain each one, which turns a weakness into evidence of careful work.
安全、伦理与环境三张清单
方法里要有一段风险评估，写明试剂、器具与操作的风险与对策。涉及人体测量时须取得知情同意，受试者可随时退出，数据匿名保存。生物实验遵守学校与考试局关于动物与人体组织的规定，野外调查需要场地许可。环境方面，废液分类回收，活体材料归还原生境。IBO 的评分说明把安全、伦理与环境列为方法学考量的一部分。漏写这一段，研究设计很难进高档。风险评估写到具体步骤，比如加热回流时用冷凝管与护目镜，才有说服力。
A short risk assessment covering reagents, glassware and procedure belongs in the method rather than an appendix. Human participants need informed consent, a right to withdraw and anonymised results; biological work follows school and exam board rules on tissue and animals. Environmental handling, from waste segregation to returning live specimens, counts as part of the methodological considerations in the IBO descriptors.
数据收集最常丢分的三处细节
阅卷时最常见的失分不在实验本身，而在交代不全。缺重复次数、缺控制手段、缺不确定度来源，这三处一出现，研究设计与数据分析就同时掉档。补法是把方法写成可复现的操作序列，任何人照着做能得到同量级的数据。表格与图各承担一半的说服力，图里的误差棒要能从表里算出来。结论只能写在数据支撑得到的范围内。数据收集做扎实，结论与评价两项才有可写的内容。报告写完之后回头对一遍评分说明的措辞，往往还能捡回一两分。
Most lost marks come from omissions rather than from poor lab work: no repeat count, no control method, no uncertainty source. Writing the method as a reproducible sequence fixes all three at once, because a reader can then repeat the numbers and see the spread. Graphs and tables each carry half of the argument, and a conclusion should never travel further than the data allows.
Cambridge International AS ALevel Physics Coursebook 正版剑桥国际AS ALevel考试物理教材带线上账号 英文原版¥557.0📱 去京东购买京东精选
需要一对一辅导？Need One-on-One Tutoring?
A-Level / GCSE / IB / KS2-KS3 全科辅导
16621398022 同微信
内部评估的选题、实验设计与数据表都可以先聊一轮，再决定怎么做。
Visit aleveler.com
或截图后用「扫一扫 → 相册」选择本图