MYP科学 Criterion C 数据处理 结论 误差 评估
MYP 科学的实验报告里，Criterion C 是唯一一条完全由数据支撑的评分项。它不考你知道多少科学知识，只看你拿到数据之后怎么处理、怎么解释、怎么评估。下面按五条 strand 拆开写，每条给出可以直接套用的写法，以及学生在报告里最常丢掉的那一分。
Criterion C sits at the point where an investigation stops being a plan and becomes evidence. The marks depend on how the data is organised, transformed and argued from, and on how honestly the method is judged.
Criterion C 的五条 strand 与四档分界线
IBO 的 MYP Sciences guide 把科学分成四个等权重的 criterion，Criterion C 叫 Processing and evaluating，满分 8 分。它管的是拿到数据以后的事：整理、变换、呈现、解释、评估假设、评估方法、写改进。五条 strand 在 Year 1、Year 3、Year 5 三张表里逐级加码，动词从 state 走到 evaluate。四档描述的分界也写得很清楚，1 到 2 分是 limited，3 到 4 分是 adequate，5 到 6 分是 substantial，7 到 8 分是 excellent。很多学生以为高分只要求数据好看，其实 7 到 8 分的 descriptor 换成 correctly collect, organize, transform and present，四个动作缺一个就掉档。IBO 还规定每个学科组每一年都要把四个 criterion 的所有 strand 至少评两次，Criterion C 不是期末才练一次的技能。
Criterion C is the processing half of an MYP science investigation, and it carries the same eight marks as knowing, designing and reflecting. The descriptors move through four bands, and the step from substantial to excellent is mostly a change of verbs: collect, organize, transform and present all have to be visible in the report, not only in the final graph. Because every strand must be assessed at least twice a year, one strong write-up will not carry the criterion on its own.
原始数据表：单位、有效数字与重复测量
数据表的骨架比外观重要。每个测量量的单位写在表头而不是每一格里，物理量用斜杠形式，例如时间 t / s，这样整列只放数字。有效数字要和仪器分辨率对齐，量筒读到 0.1 cm^3 就不要在后面补两个零充精度。重复测量要把每一次都列出来，再单独给平均值，只写平均值等于把随机误差藏起来。常见的错误是把看着别扭的读数直接删掉，改完的表看不出原始波动，后面评估方法时就没有证据。仪器本身有标称不确定度的话，写在表头或表下方，不要混进数据格。数据量大的时候把原始表放附录，正文放处理后的表，两份要能逐行对上。
A data table is judged by what a reader can extract from it without asking questions. Units belong in the header row, repeated measurements belong in separate rows, and the mean is an extra column rather than a replacement for the raw values. Hiding a reading that looks odd removes the evidence you need later when you discuss random error.
数据变换：什么时候取倒数、取对数、平方
变换是被单独列出来的一条要求，不是可选项。要不要变换取决于你想让关系变成直线：单摆的周期平方与摆长成正比，就画 T^2 对 L；玻意耳定律里 p 与 V 成反比，就画 p 对 1/V；指数衰减取半对数坐标，斜率直接给出衰减常数。变换要在表里多出一列，并注明这一列怎么算出来的，公式写在表下方。变换后的单位也跟着改，T^2 的单位是 s^2 而不是 s。很多学生把变换做在图里，正文不写公式，阅卷时看不出这是有意处理还是随手画的线。变换有一个前提，原始数据本身可靠，把噪声放大一百倍再拟合，得到的直线只会更自信地错。
Transforming data is its own strand, and the usual reason to do it is to turn a curve into a straight line whose gradient means something. A pendulum gives period squared against length; Boyle's law gives pressure against one over volume; exponential decay gives a straight line on a semi-log plot. State the transformation and its new unit in the table, otherwise the marker cannot tell deliberate processing from a careless graph.
图表：轴标签、最佳拟合线与异常点
图表的检查清单可以很短。横纵轴写清物理量与单位，刻度均匀，数据点用叉号或圆点标出，最佳拟合线穿过数据的整体趋势而不是首尾两点。异常点保留在图里，用圈标出来，再在正文里解释可能的原因。手绘和电子作图的要求不同，手绘用铅笔和直尺，电子作图不要用平滑曲线去连散点。拟合线的斜率要读出数值并带单位，截距在物理上有意义时才讨论。坐标轴不从零开始却不标注断点，是另一个高频失分点，读者会误判比例。图下面写一句图注，说明这张图在回答哪一个问题。
Graphs are read before they are admired. Label both axes with quantity and unit, plot every reading, and draw the line of best fit through the trend rather than through the first and last points. Keep outliers visible and explain them in prose; a graph with the awkward points removed looks tidy and tells the marker nothing.
从读数到不确定度：有效数字与误差传递
不确定度不是高级技巧，报告里写一句就够，但要写对。仪器读数的不确定度通常取最小刻度的二分之一，尺子读到 0.1 cm 就是正负 0.05 cm。加减运算看绝对不确定度，乘除运算看相对不确定度，最终结果的不确定度保留一位有效数字。算平均值时先算标准差，再决定小数点后留几位，别把仪器精度写得比实际高。有些量是直接测的，有些是算出来的，算出来的量必须说明用了哪几个读数。结果的规范写法是数值加单位加不确定度，例如 g = 9.7 正负 0.2 N/kg。误差传递算不下去的时候，定性说明哪一个量的不确定度贡献最大，也比什么都不写强。
Uncertainty deserves one clear sentence rather than a paragraph of hedging. Take half the smallest scale division as the reading uncertainty, combine absolute uncertainties for sums and relative ones for products, and quote the final value with a single significant figure of uncertainty. If a full propagation is beyond the course, name the measurement that contributes most and say so.
从趋势到结论：描述与解释的分工
描述和解释是两件事。描述说数据做了什么，解释说为什么会这样。写趋势时给出方向、范围和变化率，例如温度每升高 10 摄氏度，反应时间下降约 4 秒，而不是笼统地说明显下降。把结果和假设里的预测值对上，差多少要写出来，差在哪个区间也要写。用科学概念把数字接回理论：粒子运动加快、碰撞频率上升、有效碰撞比例提高，这些才是解释。比较两组数据时说明控制变量是否真的控制住了，否则差异可能来自别处。结论只写数据支持的部分，不要顺手推广到没测过的条件。
Describing a trend and explaining it are different jobs, and the 7 to 8 band asks for both. Give direction, range and rate, then connect the numbers to a mechanism such as particle motion or collision frequency. Any claim that goes beyond the conditions you tested should be marked as a prediction, not a conclusion.
假设有效性：支持、部分支持、无法判断
评估假设不是判断对错，而是说明数据支持到什么程度。三种结论都合法：支持、部分支持、不支持。部分支持最常见，因为趋势方向对但幅度不符，或者只在某一个区间成立。写的时候把假设原句引出来，再指出哪一部分被数据证实，哪一部分没有。如果数据无法判断，例如误差范围覆盖了预测值，就照实写无法区分，这比硬下结论得分高。Year 1 的 descriptor 用 prediction，Year 3 用 hypothesis，Year 5 要求 evaluate，同一份报告里提前用上更高年级的动词没有坏处。样本量小的时候，部分支持往往是唯一诚实的答案。
Evaluating a hypothesis is about degree, not verdict. Supported, partially supported and not supported are all legitimate outcomes, and partial support is the most common because a trend can point the right way while missing the predicted size. If the uncertainty range covers the predicted value, say the data cannot distinguish the two, which scores better than a confident claim the evidence does not carry.
方法有效性：随机误差、系统误差与控制变量
评估方法要落到具体的误差来源。随机误差让重复测量散开，可以靠多次测量和平均值压低；系统误差让所有读数朝同一方向偏，平均值救不了，只能靠校准或者换方法。控制变量要逐条点名：温度、光照、浓度、同一台仪器、同一个操作者。范围太窄会让趋势看起来比实际更确定，取样点太少则任意两个点就能画出一条线。评估时把误差分类再给量级，例如读数误差约正负 0.5 摄氏度，比只写有误差更有说服力。方法上的缺陷未必推翻结论，说明它限制了哪一类结论就够了。
Random error scatters repeats and can be reduced by averaging; systematic error shifts every reading the same way and survives averaging. Name the variables you held constant and the ones you could not, then put a size on the main uncertainty instead of mentioning error in general. A narrow range or three data points makes a trend look firmer than the evidence allows.
改进与延伸：写到可执行的程度
改进要针对上文自己指出的缺陷，延伸要指向新的问题。写改进时给出动作和预期效果：把秒表换成光电门，把单次测量改成五次取平均，把量程换成更小的那一档，把温度控制在正负 0.5 摄氏度以内。延伸可以改自变量的范围，例如把浓度扩到更低的一端，或者换一个物种、换一种材料重复。不要写多加几次实验这种空话，阅卷时看不出要加什么。方法本身有结构性缺陷的话，例如用肉眼判断颜色终点，改进就是换检测手段而不是多测几次。改进和延伸分开写，各给两三条，条数不多但每条都能执行。
Improvements should answer the weaknesses you have just described, and each one needs an action rather than a wish. Replacing a stopwatch with a light gate, averaging five readings instead of one, or narrowing the range of a sensor are changes a marker can picture. Extensions point forward: test a lower concentration, a different material, a wider temperature span.
Year 1 到 Year 5 的动词升级
同一份实验报告在三个年级的评分表上会得到不同分数，差别在动词。Year 1 只要求 state 和 describe，把预测和方法的有效性说出来就算达标。Year 3 换成 discuss，需要给出理由和证据。Year 5 要求 evaluate，并且 correctly collect, organize, transform and present 四个动作齐全。写报告前先确认自己用的是哪一年的 descriptor，拿错年级的表格准备会白费力气。低年级学生提前练高年级动词不吃亏，高年级学生退回 state 就会掉档。MYP 的 on-screen examination 里 Criterion B 与 C 合并占 50 分，Criterion C 的权重不低。
The same report scores differently in years 1, 3 and 5 because the descriptors ask for different verbs. Year 1 wants a stated prediction and a stated judgement of the method; year 3 wants those judgements discussed with reasons; year 5 wants them evaluated and the data properly transformed. Check which year's descriptors your school reports against before you write.
扣分点与写作顺序
阅卷时最先看的是数据和结论能不能对上，其次才看语言。扣分最多的几处很固定：表里没有单位，图里没有轴标签，结论重复数据却不解释，评估假设不引用原句，方法评估只写人为误差四个字。写作顺序建议先定表和图，再写解释，最后写评估，因为评估要引用前面的具体数字。整篇报告里每个数字都要有出处，要么来自表，要么来自计算，不能凭空出现。Criterion C 的分数集中在证据链是否完整，文字漂亮但链路断了照样停在 5 到 6 分。把草稿交给同学读一遍，让他指出哪一句看不懂，比自查三遍有效。
Most marks lost in Criterion C come from a broken chain between table, graph, conclusion and evaluation, not from plain English. A missing unit, an unlabelled axis, or an evaluation that says human error and stops there will cap the band no matter how good the writing is. Draft the table and graph first, write the interpretation against those numbers, and leave the evaluation until last so it can point at specific readings.
DK汽车大百科 升级修订版 儿童科普百科¥372.1📱 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
Visit aleveler.com
或截图后用「扫一扫 → 相册」选择本图