Dat is hoogstwaarschijnlijk wel mogelijk. SPSS heeft een zéér streng backwards compatability policy. Dit heeft als voordeel dat syntax gemaakt in versie 26 nog draait op versie 12, en dezelfde resultaten oplevert.
Nadeel is dat fouten of verouderde formules niet geupdate worden. Dalgaard (2007, p3) merkte enige tijd geleden al op dat SPSS gebruikt maakt van een verouderde, en suboptimale formule voor epsilon, in plaats van de geupdate variant van Lecoutre (1991).
The Huynh-Feldt correction (Huynh and Feldt, 1976) is
$$ \varepsilon_{HF} = \frac{(f+1)p\varepsilon_{GG}-2}{p(f-p\varepsilon_{GG})} $$
where f is the number of degrees of freedom for the empirical covariance matrix. (The original paper has N instead of f + 1 in the numerator for the split-plot design. A correction was published 15 years later (Lecoutre, 1991), but another 15 years later, this error is still present in SAS and SPSS.)
De correctie van Lecoutre (1991) is bijzonder kort; het artikel is een bladzijde lang. De relevante tekst is:
An error has occurred in the generalization of the e approximate test procedure in repeated measurements designs to the case of g > 2 groups. For a design with N subjects nested into g groups and k treatments, the formulas given in Huynh and Feldt (1976, p. 76) for the particular case of a single repeated factor and in Huynh (1978, p. 164) for the more general case of several repeated factors are, respectively:
$$ \bar{\epsilon} = \frac{N(k-1)\hat{\epsilon}-2}{(k-1)(N-g-(k-1)\hat{\epsilon})} $$
and
$$ \bar{\epsilon} = \frac{Nr\hat{\epsilon}-2}{r(N-g-r\hat{\epsilon})} $$
They should be corrected, when g > 2, by substituting N - g + 1 for N in the numerator. Hence, the correct formula should be in the general case:
$$ \bar{\epsilon} = \frac{(N-g+1)r\hat{\epsilon}-2}{r(N-g-r\hat{\epsilon})} $$
The above incorrect $\bar{\epsilon}$ is mentioned in many recent textbooks and is apparently used in a number of standard statistical packages. It is at the present time a routine solution when the condition of circularity (or sphericity) is not fulfilled. It gives an underestimation of the deviation from circularity, which may be substantial when the total number of subject N is small.
Dit betekent dat je meer kunt vertrouwen op JAMOVI-output betreffende epsilon-estimates, dan op SPSS.