Check effect size conversions

[LukasWallrich]

Currently, a fair number of effect sizes cannot be converted - @LukasRoeseler, can you check again if any of the below ES types can be converted (or changed to a more common spelling in the data - particularly for those that do not really exist such as cohen's g)

  Unknown effect size types: dz, hazards ratio, w, cramer’s v, beta,  , b, 
    etaq, γ, q, cramer's v, regression coefficient, cohen's g, percentage, φ2, ges, 
    χ2, h, odds ratio (study 3), unstandardized coefficient
  Unknown effect size types: hazards ratio, w, dz, beta, b, q, cramer's v, 
      squared semi partial correlation (sr2), percentage, ges, etaq, χ2, h, unstandardized coefficient

[LukasRoeseler]

Listing changes and remarks for discussions here:

• Cohen's dz: no changes - dz can be converted to d if the correlation between the two vectors is known (if r = .5, dz = d)
• harzards ratio: no change - I could not find an existing solution to convert from HR to r and more information may be necessary (see for example https://www.researchgate.net/publication/6359972_Transforming_Odds_Ratios_Into_Correlations_for_Meta-Analytic_Research).
• w: no change - This concerns a replication by Genschow (door in the face). I could not find the statistics in the article and thus could not replace them with other value as the authors probably calculated the effect sizes themselves. I reached out to them for clarification and highlighted the entry. I did not find a way to convert from W to another statistic.
• Cramer's V: no change - applies to crosstables that exceed 2x2
• beta: no change - could be interpreted like a correlation but only if it is the only predictor in a model
• b: no change (see regression coefficient)
• etaq: corrected 4 instances of etaq in the dataset, also the function "estype_map" now recognizes etaq as etasq
• regression coefficient: no change, cannot be converted to correlation unless it is for an only predictor and further statistics are known
• cohen's g: found 1 instance (not yet validated entry), checked original article and found that original effect size was unclear. Effect size was removed. Changes are marked in "notes_validation".
• percentages: no change
• phi2: Original article reported phi² (CORE data, unvalidated). I could compute a phi value from the reported frequencies in the original paper and replaced phi² with phi in the dataset. Changes are marked in "notes_validation".
• ges: no change - One instance was found (CORE data, unvalidated). I did not check where "ges" comes from as the replication paper entry was incorrect. I made a remark in "reason for exclusion".
• χ2: no change - I could not find any cell with this (except for test-statistics cells for XPHI entries but these already have coded correlations).
• unstandardized coefficient: see regression coefficient
• squared semi partial correlation (sr2): no change - due to the correlation being partial, a conversion is not possible without further info
• γ (gamma): no change - These are coefficients from a linear model (Vaughn et al., 2023, rows 1843-1852). Conversion is not possible.
• q: no change - concerns ML2 rows 1 and 2 (1875-1876). I could not find a way to convert from Cohen's q (and neither did Klein et al. as they used two separate plots for the studies using q vs. those that could be converted to r).

Not yet closed:

• odds ratio (study 3): I could not find an instance with this effect size. @LukasWallrich, can you please specify the variable name? For now, I made the "estype_map" function recognize this as "or".
• h: no change - I could not find the entry, @LukasWallrich, can you please provide further info about this?

[LukasWallrich]

@LukasRoeseler the effects you could not find are in FORRT. I removed the (study 3) bit from the OR, so that is fine - but could you look at the h values? I have no idea what they are ...

[LukasWallrich]

Here the explanation of h - not sure if esc has the conversion built in - https://chatgpt.com/share/eda14b35-fbb3-4a2c-a612-cb471a642944