Uni Reading List
Study guides /

Psychological research: distinguish a design from a complete analysis

An experiment’s assignment rule, measurement and available outcomes are different parts of its evidence. This route asks which claim a design can support and how missing scores change the calculation. The goal is an auditable argument, not a button sequence or a claim that a significant result repairs the design.

Course topic: Research Methods and Statistics

Study route and exercise: University Reading List editorial project. How we curate · Sources checked:

Download blank worksheet (Markdown) ↓
On this page

Scope and responsibility

For introductory psychological methods with arithmetic, means, proportions and basic graph reading. All units, scores and assignments below are synthetic. No study is conducted and no human data are used. The calculations describe finite assigned-arm means and explicit imputations; they do not by themselves identify a population causal effect.

Choose your starting point

I need a free broad methods spine

Use Jhangiani and colleagues for measurement, designs and basic statistics. Practise distinguishing manipulation, measurement and sampling before selecting a test.

I struggle to evaluate published claims

Use Morling’s consumer/producer framework to connect claim type with evidence and validity. A well-labeled claim still needs outcome completeness and uncertainty checks.

I need help interpreting numerical analysis

Use Reidy and Dancey as the statistics companion after specifying the design. Software output cannot decide what the missing observations would have been.

Compare the three books

Research Methods in Psychology

Rajiv S. Jhangiani; I-Chant A. Chiang; Carrie Cuttler; Dana C. Leighton · Edition: 4 · 2019

Free design-and-measurement spine

Choose it when
You need an accessible complete methods course with experimental and non-experimental alternatives.
Use another route when
You want free access to mean unrestricted commercial reuse, or need a specialist missing-data monograph.
Selected reading
2019 fourth edition: Psychological Measurement, Experimental Research, Non-Experimental Research, then Descriptive and Inferential Statistics.
Limits and trade-offs
The author-hosted book is free under noncommercial/share-alike conditions, with third-party exceptions. The selected ISBN identifies the e-book; the print identifier differs.

Try before choosing

Write the manipulation, measured outcome, assignment unit and missingness count for the original packet.

Check your work: Random assignment, random sampling and complete observation are three different properties.

Official free reading ↗

Book record and access options →

Research Methods in Psychology: Evaluating a World of Information

Beth Morling · Edition: 5 · 2025

Claims-and-validity spine

Choose it when
You want to interrogate the relation between a reported claim and its design.
Use another route when
You need its framework to substitute for checking the actual sample, measurement or missing-data assumptions.
Selected reading
Fifth edition, 2025: chapter 3 on claims/validities,5 measurement,7 sampling,10–11 experiments; use the limited publisher demo only for its stated sample access.
Limits and trade-offs
The consumer/producer activities contrast with a test-selection manual. Commercial courseware and the complete book are not made free by the existence of a demo.

Try before choosing

Write the strongest defensible claim from complete-case means, then a stronger claim that requires an extra assumption.

Check your work: Name the exact extra assumption. An initially randomized allocation does not reveal scores subsequently missing.

Book record and access options →

Statistics without Maths for Psychology

John Reidy; Christine P. Dancey · Edition: 9 · 2024

Statistics-and-reporting companion

Choose it when
You need verbal and graphical explanations of common analyses and their interpretation.
Use another route when
You expect a software recipe or p-value to settle design validity or selection into observation.
Selected reading
Ninth edition, 2024 paperback ISBN 9781292463445: chapter 1 design,3 descriptives,4 sampling,7 two-condition comparisons,8 issues of significance.
Limits and trade-offs
Its SPSS guidance is useful when a course uses that software, but a book or eText subscription does not itself supply an SPSS licence. The reported results must answer a defined estimand.

Try before choosing

Compute observed means and their denominators before considering any significance test.

Check your work: Report assigned n, observed n and missing n separately, with score units. A mean without its population is ambiguous.

Book record and access options →

Check your preparation

What is the target quantity?

Here it is the difference between mean scores of all ten assigned units in each arm, not merely completers. A causal population quantity would require additional design and inferential reasoning.

What does random assignment solve?

It specifies how initial groups were allocated. It does not automatically make later missingness random, measurements valid or the synthetic sample representative.

What is a sensitivity assumption?

A transparent restriction on unknown values used to see how a conclusion changes. It is not an observed fact or a licence to fill values without reporting the restriction.

A six-step study plan

These are your own study marks, not an assessment. They are not sent to this site.

  1. Define the claim and unit

    Selected reading: Morling claims discussion or Jhangiani design overview.

    What to do: Write the outcome, unit, assignment mechanism and target group.

    Check your work: Do not slide from an observed-arm contrast to an unqualified population effect.

  2. Audit observation

    Selected reading: Experimental-validity reading.

    What to do: Make assigned/observed/missing tables for both arms.

    Check your work: Attrition rates use assigned denominators, not only those who supplied a score.

  3. Calculate complete cases

    Selected reading: Reidy/Dancey descriptives or the free statistics part.

    What to do: Compute sums, means and the observed contrast with units.

    Check your work: The denominator is visible and the result is labeled complete-case.

  4. Bound what is unknown

    Selected reading: Measurement limits and the original example.

    What to do: Use the stipulated 0–10 scale to obtain the lowest and highest assigned-arm contrast.

    Check your work: For a lower difference, minimize A’s missing values and maximize B’s; reverse for the upper bound.

  5. Add one explicit assumption

    Selected reading: The changed-assumption follow-up.

    What to do: Set a common missing-score mean m and derive the contrast as a function of m.

    Check your work: This is not the same as setting each arm’s missing mean equal to its observed mean.

  6. Write a reproducible conclusion

    Selected reading: Reporting discussion in the chosen spine.

    What to do: Report observed contrast, unconstrained bound and conditional sensitivity result together.

    Check your work: Separate arithmetic uncertainty from sampling uncertainty and from whether the outcome measures the intended construct.

Worked exercise

Original attrition packet: the observed contrast can reverse

Synthetic arms A and B each start with ten assigned units. A has six observed 0–10 scores: 8,8,9,9,10,10. B has eight: 5,5,6,6,7,7,8,8. Other scores are missing but must lie in [0,10]. Compute complete-case means and the possible difference in all-assigned means A−B.

Assumptions

Initial allocation is stipulated randomized, but nothing is known about why scores are missing. The score is in points, bounded inclusively by 0 and 10. Every assigned unit has one underlying score. Treat this as a finite arithmetic packet, with no population or causal-effect claim supplied by the numbers.

  1. A’s sum is 54 and observed mean 54/6=9; B’s sum 52 and mean 52/8=6.5. Complete-case contrast is 2.5 points. Attrition is 4/10=40% versus 2/10=20%.
  2. All-assigned A lies between 54/10=5.4 and (54+40)/10=9.4. B lies between 52/10=5.2 and (52+20)/10=7.2.
  3. The smallest A−B is 5.4−7.2=−1.8, attained by A’s missing scores all 0 and B’s all 10. The largest is 9.4−5.2=4.2, with those choices reversed.
  4. Both signs are compatible with the supplied observations. Randomization of assignment does not choose between these missing-value completions, and no p-value from the observed cases supplies them.

Answer and interpretation

Observed means 9 versus 6.5, contrast 2.5 points; all-assigned contrast can range from−1.8 to 4.2 points. The observed positive sign is not secured for the stated finite target.

Test the result

Construct both endpoint completions explicitly and keep ten as each target denominator. The bound concerns missing outcomes, not a confidence interval or an estimate of a population causal effect.

Change an assumption

Changed assumption: equal missing means across arms

Keep all observed values but assume A’s four missing scores and B’s two missing scores share the same mean m, with 0≤m≤10. Derive the all-assigned contrast, its range and its value at m=6. This is an assumption, not a finding about the missingness mechanism.

  1. A’s assigned mean is (54+4 m)/10=5.4+0.4 m; B’s is (52+2 m)/10=5.2+0.2 m. Their difference is 0.2+0.2 m.
  2. As m ranges 0…10, the conditional difference ranges 0.2…2.2 points and remains positive. At m=6, A=7.8, B=6.4 and A−B=1.4.
  3. The common-mean restriction removes the negative completions from the main packet. It is not implied by random assignment and differs from imputing 9 to A and 6.5 to B. No sampling uncertainty or population effect has been settled.

Answer and interpretation

Under the common missing-mean assumption only, the finite contrast is 0.2+0.2 m, positive over [0,10]; at m = 6 it is 1.4 points.

Test the result

Withdraw the common-mean restriction and the original−1.8…4.2 bound returns. Report both results so that a reader can see which conclusion depends on the assumption.

Common mistakes

Randomized means complete

Random allocation does not recover an absent outcome or establish representativeness.

Unknown means zero

Zero is one possible score, not the default meaning of a missing record.

Sensitivity result presented as observed

An imputation rule changes the admissible values; it must appear beside the resulting claim.

Before moving on

  • I can define the unit, target mean and observed denominator.
  • I can reproduce the missing-score endpoints and common-mean sensitivity formula.
  • I can distinguish a finite contrast, a conditional calculation and an inferential causal claim.

Sources and edition checks

Sources checked:

These are proposed study activities, not official course assignments or a claim that every book has been read cover to cover.

How we curate →