Psychométrie et construction d'épreuves
GRE Subject Test · GRE Psychology Sujet 10 4:58 Narration en anglais · Sous-titres anglais + 中文 incrustés
Chapitres
Transcript
Choose the reliability coefficient by the source of variation being studied.
根据正在研究的变异来源选择信度系数。
A standard error of measurement concerns measurement uncertainty under a model; it is not the standard error of a sample mean.
测量标准误涉及模型下的测量不确定性，并非样本均值的标准误。
The Spearman–Brown estimate assumes sufficiently comparable halves; a formula cannot repair unrelated item content.
斯皮尔曼-布朗估计假设两半测验具有可比性；公式无法弥补项目内容的不相关。
Standard scores describe relative position and do not show that a test measures the intended construct fairly across groups.
标准分数描述相对位置，不能表明测验在各群体间对目标构念的测量是公平的。
Reliability is consistency, and its type is named by the design that produced it.
信度即一致性，其类型由产生该信度的设计命名。
The same test twice gives test-retest reliability.
同一测试施测两次得到重测信度。
Two similar forms give alternate-forms reliability.
两个平行形式得到复本信度。
Consistency within one sitting, such as split-half or Cronbach alpha, is internal consistency.
单次施测内的一致性，如分半法或克朗巴赫α系数，属于内部一致性。
Agreement between scorers is inter-rater reliability.
评分者间的一致性是评分者信度。
The Spearman-Brown formula r sub full equals 2 r divided by 1 plus r rescales a half-test correlation to full test length.
斯皮尔曼-布朗公式 r_full = 2r / (1+r) 将半测验相关系数重新缩放至全测验长度。
For example, if r equals 0.80, it becomes about 0.89.
例如，若 r = 0.80，则结果约为 0.89。
Validity is evidence for the proposed interpretation, not of the test itself: content validity asks whether items cover the domain; criterion validity relates scores to an outcome, concurrently or predictively; construct validity assembles convergent and discriminant evidence that the score tracks the intended trait. Reliability caps validity but never establishes it: a consistent scale can systematically measure something else.
效度是针对所提议的解释的证据，而非针对测验本身。 内容效度询问项目是否覆盖了该领域。 标准效度将分数与结果相关联，包括同时效度和预测效度。 构念效度整合了聚合效度和区分效度的证据，以表明分数能够追踪目标特质。 可靠性限制了效度的上限，但从未确立效度。 一个一致的量表可能系统性地测量其他东西。
Standard scores re-express raw scores against a norm group.
标准分数把原始分数与常模群体相比较重新表示。
z equals X minus the mean, over the standard deviation; T equals 50 plus 10 z; and the Wechsler deviation I Q equals 100 plus 15 z.
z 等于 X 减去均值，再除以标准差；T 等于 50 加 10z；韦氏离差智商等于 100 加 15z。
A z of plus 2 gives an I Q of 130, about the 98th percentile.
z 为正 2 时，智商为 130，约为第 98 百分位。
The standard error of measurement equals the standard deviation times the square root of 1 minus r, which converts reliability into a score band; a standard deviation of 15 with r equal to 0.91 gives 15 times the square root of 0.09, which is 4.5.
测量标准误等于标准差乘以 1 减 r 的平方根，它把信度转化为分数区间：标准差为 15、r 为 0.91 时，等于 15 乘以 0.09 的平方根，即 4.5。
Item analysis checks each item; difficulty p is the proportion answering correctly, and discrimination is the item-total correlation.
项目分析逐题检查：难度 p 是答对的比例，区分度是题目与总分的相关。
Items that nearly everyone or no one answers correctly carry little information; negative discrimination flags a likely miskeyed or misleading item.
几乎人人答对或无人答对的题目信息量很少；负的区分度提示该题可能答案键有误或表述误导。
A fictional scale has mean 50 points, SD 10 points and reliability 0.84.
一个虚构的量表均值为50分，标准差为10分，信度为0.84。
A score is 65.
某得分为65分。
Compute z and SEM, showing the formulas.
请计算z分数和SEM，并展示公式。
Z equals X minus mean, all over SD.
Z等于X减均值，除以标准差。
Z equals 65 points minus 50 points, all over 10 points, which equals 1.5, dimensionless.
Z等于65分减50分，除以10分，结果为1.5，无量纲。
SEM equals SD times the square root of 1 minus r.
SEM等于标准差乘以根号下1减r。
SEM equals 10 points times the square root of 1 minus 0.84, which equals 4 points.
SEM等于10分乘以根号下1减0.84，结果为4分。
SEM is a model-based uncertainty measure, not proof of validity.
SEM是基于模型的测量不确定度指标，并非有效性的证明。
T equals 50 plus 10 times z.
T等于50加10乘z。
T equals 50 plus 10 times 1.5, which equals 65 on the T scale.
T等于50加10乘1.5，在T量表上结果为65。
Tight grouping is like reliability, being on target like validity: a test can be reliable but not valid.
弹着点集中好比信度，命中靶心好比效度：测验可以可靠却不有效。
Read the labels before drawing a conclusion.
先读清标签，再下结论。
Two raters repeatedly disagree by five points but rank scripts identically.
两位评分者反复出现五分差距，但排名完全一致。
Does a high correlation establish agreement?
高相关系数能否证明一致性？
Suggest a check.
请提出一种检验方法。
Pause before the model.
请先暂停，完成这道原创练习再继续。
No.
不。
Rank association can remain perfect with systematic score differences.
即使存在系统性的分数差异，等级关联仍可能保持完美。
Inspect paired differences and an agreement measure suited to the use, with a clear rubric and rater training.
请检查配对差异，并使用适合该用途的一致性度量，同时需具备清晰的评分标准和评分员培训。
Correlation alone does not establish interchangeable scores.
仅凭相关系数无法确立分数的可互换性。
Distinguish consistency from absolute agreement.
请注意区分一致性（consistency）与绝对一致性（absolute agreement）。
Calling a two-different-forms correlation test-retest reliability, or citing high reliability as if it were validity evidence.
将两种不同形式的分数相关称为重测信度，或将高信度当作效度证据来引用。
State the source and the limits.
请说明数据来源及局限性。
A highly reliable score predicts poorly in a new population.
一个高度可靠的分数在新人群中预测效果很差。
Explain why reliability did not guarantee this validity claim.
请解释为什么可靠性未能保证这一效度主张。
Pause and write a justified answer.
请先暂停，完成这道原创练习再继续。