FIGSHARE:24949470

This raw dataset provides n = 134,699 individual Likert-scale responses to a series of 39 policy issues and value statements along with additional user demographics and preferences for most of the responses

Data Records, User Responses Dataset; Table 1

The raw file has 134,699 rows and every column of Table 1 and the data dictionary (69 columns). 81.0% of rows give demographics, 67.6% at least one salient statement, 73.2% a party considered and 76.9% a party never considered. 133,473 rows remain after removing those with -1, 133,143 after the R code's cleaning.

This dataset lists the n = 39 statements used in the survey across three languages — Slovak, Hungarian, and English and offers detailed information on various parties’ positions concerning specific statements with n = 429 observations

Data Records, Party Positions Dataset

429 coded rows on 39 statements, each with one Slovak, one Hungarian and one English wording. Every row has expert positions of both coders and a final position; 131 rows have a self-placement. Snippet and link columns are present for self-placements, each coder and the final position. The "Self-placement source" column is empty in all rows.

How the test differed from the paper’s method

• Statements are numbered by their English wording, matched to Supplementary Table 1, since the workbook's "Statement no." is wrong for 99 rows (see findings). Both numberings give 39 statements.

Out of 25 political parties, 11 met the inclusion criteria and were incorporated into our VAA

Methods, Party selection and coding, paragraph 1

The workbook names 11 parties (ALIANCIA, DEMOKRATI, HLAS-SD, KDH, OLANO, PS, Republika, SME Rodina, SMER-SD, SNS, SaS), each with a position on all 39 statements; the recoded CSV holds the same 11.

Responses were garnered from four political entities out of eleven

Methods, Party selection and coding, paragraph 4

Four parties have self-placements, SaS on 39 statements, DEMOKRATI on 38, PS on 32 and KDH on 22. PS used only "agree" and "disagree" (19 and 13), as a note in the workbook says.

we employed Weighted Cohen’s Kappa37, which provided a value of κ = 0.84, indicating a high level of consistency among coders

Methods, Party selection and coding, paragraph 3

The workbook holds each coder's positions. With quadratic weights and pairs where either coder chose "No opinion" left out (266 of 429), kappa is 0.842. Linear weights give 0.734 (No opinion left out) and 0.620 (No opinion as Neutral); quadratic weights with No opinion as Neutral give 0.745. Only the first reading is within 0.84 +- 0.084. The paper names neither the weights nor the handling of No opinion; this is asked of the authors (findings).

How the test differed from the paper’s method

• Assumed quadratic weights with "No opinion" pairs left out, the one reading of the four the criterion names that gives the reported value; the paper says only "Weighted Cohen's Kappa" (Methods, Party selection and coding, paragraph 3). Quadratic weights are the common choice for ordinal scales, and No opinion has no place on the ordered scale, so this reading is the best supported; all four are shown.

We used the ‘recoded_party_positions’ dataset and computed pairwise Manhattan distances to emphasize the absolute differences in policy positions, applying average linkage clustering

Technical Validation, Content validation, paragraph 7; Fig. 7

With the recoded positions (No opinion as 50, reading a), Manhattan distances and average linkage, all 9 non-trivial clusters of Fig. 7 are found and no other, at heights within 1 of those read from the figure (root 1774.0, figure 1774). The figure was read blind to the computed tree, by pixel position against its axis ticks. Reading (b), No opinion left out of each distance, finds 7 of 9 clusters; the variants on coder I's and coder II's positions find 2 and 7 of 9. The deposited R code clusters the recoded CSV, as reading (a) does.

Not quoted: The numbers are only in Table 2; no sentence of the text states them.

Every reading is within tolerance. Reading (a), the paper's cleaning with gender given, gives 44,938, 61,670 and 1,455 of 108,063 (largest difference 2.5%). Gender, age and education are empty in the same 25,639 rows, so readings (c) and (d) equal (a) and (b). The R code's cleaning, which also drops 330 rows that give one answer to all 39 statements, gives Table 2 exactly (44,919, 61,626, 1,420 of 107,965).

the substantial numbers within the younger age brackets, particularly those aged 25–34 (n = 34,894, 32.32%) and 35–44 (n = 27,847, 25.79%), highlight this trend (Table 2)

Technical Validation, Content validation, paragraph 2; Table 2, Age rows

Reading (a) gives every count within 2.4% of Table 2 and the two percentages within 0.01 point; every reading is within tolerance. The R code's cleaning gives every age count of Table 2 exactly.

The pronounced presence of individuals with university degrees or academic higher education (n = 69,542, 64.41%)

Technical Validation, Content validation, paragraph 2; Table 2, Education rows

Reading (a) gives every count within 2.6% of Table 2 (no education 1,015 against 990); every reading is within tolerance. The R code's cleaning gives every education count exactly.

S2 — ‘People who earn higher wages should have a higher income tax rate’ (mean = 43, s.d. = 33, median = 25) and S3— ‘The retirement age should rise as life expectancy increases’ (mean = 42, s.d. = 33, median = 25) highlight the diversity of opinions

Technical Validation, Content validation, paragraph 4; Fig. 5

On the cleaned rows, with No opinion left out, S2 has mean 43.77, s.d. 33.82, median 25 (130,235 answers) and S3 mean 42.09, s.d. 33.69, median 25. On the raw rows the S2 median is 50, outside the tolerance, and the other five values are within it.

S4 — ‘In the health care system, more emphasis should be placed on the personal responsibility of the individual’ (mean = 70, s.d. = 26, median = 75,) and S5 —‘Banks and big corporations should be taxed more heavily’ (mean = 69, s.d. = 29, median = 75) suggest a stronger consensus ... more than n = 17,435 for S4 and n = 23,344 for S5 — diverging on these views, with scores of 0 and 25

Technical Validation, Content validation, paragraph 4; Fig. 5

On the cleaned rows S4 has mean 70.36, s.d. 26.58, median 75 and 17,460 answers of 0 or 25, and S5 mean 69.38, s.d. 29.53, median 75 and 23,369, so the counts are within tolerance (reading i) and above the paper's (reading ii). The raw rows also hold. The R code's cleaning gives exactly 17,435 and 23,344, the paper's numbers, so "more than" in the paper reads as "as many as" for that selection.

Both parallel analyses indicated that 11 factors should be extracted.

Technical Validation, Content validation, paragraphs 5 and 6; Fig. 6

On the cleaned rows with No opinion as 50 (reading a), parallel analysis with psych's defaults (minres, 20 iterations) gives 11 factors for both the polychoric and the Pearson matrix. The 11th Pearson factor value is 0.030 against a reference of 0.013 and the 12th 0.004 against 0.012. The 12th polychoric value, 0.0106, is close to the normal-data reference (0.0117), but psych's own reference for cor poly, polychoric correlations of resampled data, gives 0.0143 (s.d. 0.0009 over 20 iterations) and again 11. The 95th percentile of each reference gives the same counts. Reading (b), No opinion missing, gives 12 (Pearson) and 11 (polychoric, against either reference) with the defaults, so with the defaults the claim holds under reading (a) only. With the R code's settings (ML extraction, 1000 iterations for Pearson and 100 for polychoric, on a correlation matrix, for which psych simulates normal data) every row reading gives 11 and 11, including the code's unique-email rows (29,493). 18 of 2,040 uniqueness fits on simulated reference matrices did not report convergence; every fit on an observed matrix (12) and on resampled data (40) converged.

How the test differed from the paper’s method

• The parallel analysis is a reimplementation of psych's fa.parallel in Python (factor values of the correlation matrix with one-factor communalities on the diagonal, compared with the mean over simulated normal data, and for the polychoric matrix under the defaults also over resampled data with polychoric correlations, as psych does for cor poly on raw data), not psych itself, which cannot be installed in the entry's Python environment.
• The R code's settings (fm "ml", n.iter 1000 and 100, unique-email rows) are shown as variants, taken from Volebny_Kompas_code.R (section "CORRELATION AND FACTOR ANALYSIS"); the criterion names psych's defaults.
• Polychoric correlations use thresholds from each statement's margins without psych's correction for empty cells.

gathered during the 26 days leading up to and including the day of the 2023 snap Slovak Parliamentary Elections

Background & Summary, paragraph 3; Methods, Volebný Kompas implementation and usage, paragraph 5

The first response is 2023-09-05 11:26:31 and responses cover all 26 days to 30 September, but 432 of 134,699 rows (0.32%) are dated 1 to 9 October 2023 as given, and 494 (0.37%) when TIME.STAMP is read as UTC and converted to Slovak time; both readings cover all 26 days. The criterion requires every timestamp inside the window under one reading, and it fails under both; 99.7% of rows are inside it, so the claim holds in part. The file is sorted newest first, so the sample entry of Table 1 (2023-10-09 08:57:38) is its first row and the latest timestamp. The R code has an optional filter ending at 2023-09-30 22:42:30, which drops 451 rows of data_cleaned_corrected, the rows it is applied to (461 of the raw rows). The code parses the timestamps and the cut-off with tz "UTC", so it compares the values as given, and its comment gives the end of the moratorium as 22:45 with no time zone; this suggests local time (reading a) but does not settle it. The outcome is the same under both readings.

Out of the 39 statements, 18 were assigned to the five uni-dimensions ... Of the 39 statements, 16 influenced this two-dimensional political landscape

Methods, Volebný Kompas implementation and usage, paragraphs 3 and 4; Figs. 2 and 3; Technical Validation, Content validation, last paragraph

Supplementary Table 1, which is in the dataset, assigns 18 statements to the five uni-dimensions and 16 to the landscape (11 economic, 6 cultural, statement 10 on both). With No opinion as 50 (reading ii), every party's score on each of the five uni-dimensions is within 0.8 of Fig. 2, which places parties at whole percentages, and its cultural score is within 0.3 of Fig. 3. On the economic axis 10 of 11 parties are within 0.2 of Fig. 3, but PS scores 47.7 where Fig. 3 places it at 54.5, 38 pixels right of the centre line, so PS and DEMOKRATI (52.3, figure 52.2) are ordered differently. Leaving out statement 10 (on both axes) or 21 does not explain it; PS then scores 52.5 or 47.5, still below DEMOKRATI (55.0), and other parties move away from the figure. With No opinion left out (reading i) four of the five uni-dimensions differ from Fig. 2. The order matches on six of the seven axes, so the claim holds in part. The figure values were read blind to the computed scores, by pixel position (round 1 of verification asked for this). The paper's statement that the placements conform to anticipated placements is a judgement the data cannot test; the criterion tests agreement with the figures. The dimension assignments come from the dataset's own Supplementary Table 1, so no outside labels and no link claim are needed.

How the test differed from the paper’s method

• The expected placements were read from the published Figures 2 and 3 by a separate agent that had not seen the computed scores, from each logo's centre in pixels interpolated between the axis ends (Fig. 2 to about 0.3, Fig. 3 to about 0.3 to 1). A party matches within 1 point, and a pair is ordered differently only when the figure and the scores order it opposite ways by more than 1 point each.
• Readings (a) and (b) of the criterion give the same number for a mean of polarity-adjusted positions, so only one is computed.
• Statements are numbered by their wording, not by the workbook's "Statement no.", which is wrong for statements 16 to 24.

specifically n = 40 cases under the age of 18 and n = 400 cases aged ‘18–20’, which were not congruent with university-level education attainment

Technical Validation, Structural validation, paragraph 3

With the university category only, 27 respondents under 18 and 378 aged 18-20 on the cleaned rows, and 27 and 382 on the raw rows; with applied higher education added, 34 and 505, and 34 and 509; the R code's cleaning gives 25 and 375, or 32 and 500. The 18-20 count is within tolerance under reading (i), but no reading brings the under-18 count within 4 of 40, so one of the two counts holds. The deposited R code's validation rules check only that AGE and EDUCATION take valid categories; no rule relates them. The paper says that upon initial review the team broadened the acceptable age range for a university degree to include 21-24, and that the 40 and 400 discrepancies "were acknowledged during further data curation and refinement of the data collection tool", so the counts may have come from an earlier state of the data; this is asked of the authors (findings) and of a reader (questions/figshare-24949470.yaml).

various estimation strategies were employed, including Maximum Likelihood (ML) and principal axis factoring, with promax rotation. This approach yielded consistent results. Notably statements such as S1 — ‘Real estate tax should be increased to lower labor taxes rises’ and S3 — ‘The retirement age should rise as life expectancy rises’ had strong loadings

Technical Validation, Content validation, paragraph 6

On the Pearson matrix of the cleaned rows (reading a), S1 has a largest loading of 0.523 (ML) and 0.502 (PA) and S3 0.598 and 0.603, and 35 of 39 statements load most on matched factors (lowest Tucker congruence 0.892), so the claim holds there at both thresholds; reading (b) gives S1 0.553 and 0.539, S3 0.598 and 0.603, 35 of 39. On the polychoric matrices S1 and S3 load above 0.5 but only 32 of 39 statements match (lowest congruence 0.355 and 0.358), so the solutions are not consistent there. The criterion takes the matrices of efa-eleven-factors, Pearson and polychoric, so the claim holds on one of the two, in part. The R code runs ML and PA with promax only on the Pearson matrix; on its unique-email rows the Pearson PA loading of S1 is 0.379 and 33 of 39 match.

How the test differed from the paper’s method

• The thresholds for "strong" (0.4 or 0.5) and "consistent" (35 of 39) are the claims file's own; the paper gives none, so the outcome rests on them.
• ML, principal axis and promax are reimplemented in Python after psych and stats::promax.