Specification Menu

Task D2 – choosing your analytical path

The specification menu defines the space of justifiable analyses for Task D2. Every option is defensible – but you must write the defence. That is the exercise.

The menu mirrors the structure of Gelman’s ‘garden of forking paths’: at each decision point, there is more than one reasonable choice, and choosing without reasons is what turns a multiverse into noise.

The menu

Axis	Options (choose one per axis)
Outcome	`mcosmo` (cosmopolitan framing – the Multi100 anchor) · `mutil` (utilitarian) · `mcomm` (communitarian) · `mlib` (libertarian) · `mpos` (all positive framings – closer to the claim’s “positive dimensions” wording) · `mneg` (all negative framings)
Exposure form	raw `unemp` · mean-centred `unemp_c` · `log(unemp)` · log-transformed and z-standardised
Second regressor	none · `+ growth` (GDP growth rate – is it a confounder or a co-exposure?)
Estimator	two-way FE via `fixest::feols(... \| cntry + year)` or `plm(..., effect = "twoways", model = "within")` · country FE only · random effects (`plm(..., model = "random")`) · pooled OLS with cluster-robust SE
Sample	all years (2004–2013) · pre-crisis only (2004–2008) · crisis years only (2009–2013) · exclude bailout recipients · exclude Greece and Spain (extreme unemployment outliers)
Weights	none · `n_cy` (country-year respondent counts)

With six axes and the options above, well over a thousand combinations exist – the pre-computed grid behind the closing chart holds 1,680 of them. You are choosing one. Your classmates are choosing others. Together you are sampling the multiverse.

Worked example

Choice: outcome = mpos, exposure = raw unemp, second regressor = none, estimator = two-way FE via feols, sample = all years, weights = none.

Justification (2 sentences): The ‘positive dimensions’ wording in the claim maps more directly onto the composite mpos scale than onto mcosmo alone, so using mpos is the more literal operationalisation of the stated hypothesis. Grand-mean centring the unemployment rate changes the intercept but not the slope when fixed effects are present, so using raw unemp produces an identical coefficient; the choice is made explicit here to avoid an undocumented transformation.

This is the kind of justification that makes a specification defensible: it connects the analytical choice to the claim’s wording or to a causal argument, rather than to convenience or preference.

Preregistering your choice

Commit to your specification before you run it – that is what stops you picking the path that happens to give the answer you want. How you record the commitment depends on your track, but the gesture is the same.

Track B (git): commit your justification before you add the model-fitting code. The commit message format is:

prereg: <DV> ~ <IV> | <estimator> | <sample>

Examples:

prereg: mpos ~ unemp | twoway FE | all years
prereg: mcosmo ~ log(unemp) + growth | country FE | 2004-2008
prereg: mutil ~ unemp_c | pooled OLS cluster-robust | excl GR+ES

The timestamp on the commit is your preregistration. Write the justification in your report file (the Task D2 callout block) and commit it together with any setup code, before the model-fitting code.

Track A (browser): write your chosen spec and a one-sentence reason into your script as a comment, and tell your partner – before you run. That is your commitment; you do not need git for it.

Reporting your result

After running your model, submit your result to the class-results Google Form. You do not fill it in by hand: the report_result() helper prints a one-click link with all your numbers already in it – open it and press Send. Your result then appears on the live Multiverse chart.

The helper standardises every submission to a partial correlation, r = t / sqrt(t^2 + df) – the same conversion Multi100 used – so that models on different outcome scales and estimators all sit on one comparable axis.

Track A (browser): the helper is already on the Browser lab page. Track B (template): it lives in your template repository at R/report_result.R:

source("R/report_result.R")
report_result(my_model, spec = "mcosmo ~ unemp_c | twoway FE | all years")

This prints your submission link, ready to click.

Notes on specific choices

Outcome variables. The four Teney dimensions (mcosmo, mutil, mcomm, mlib) are 0–1 relative-frequency scales. mpos and mneg are composites (means of the positive and negative sub-dimensions respectively). All are in teney_panel.csv; check the codebook (data/teney_panel_codebook.md in the template repo) for exact definitions.

Exposure form. Log-transforming unemployment makes sense if you believe the effect is proportional rather than additive; centring or standardising makes coefficients more comparable across specifications but does not change the t-statistic. Document your choice.

GDP growth as a second regressor. From the DAG perspective: is growth a confounder (a common cause of unemployment and EU framing), a co-exposure (another pathway you want to estimate separately), or a mediator (part of the path from unemployment to framing)? Your answer determines whether including it is theoretically defensible for your estimand.

Estimator choice. Two-way FE controls for time-constant country traits and year-specific EU-wide shocks. Country FE only removes country traits but leaves year shocks in the error. Random effects assumes the country-level heterogeneity is uncorrelated with the covariates – testable via a Hausman test. Pooled OLS with cluster-robust SE ignores the panel structure entirely; it requires the strongest assumption. The choice is consequential.

Sample restrictions. Dropping Greece and Spain removes the two countries with unemployment rates far above the rest during the crisis period. If the effect is driven mainly by these outliers, the coefficient will shrink substantially. That is a legitimate sensitivity check.