I do not recommend the approach used in the linked study. Truth be told, I cannot figure out why the authors used the statistical procedures they did, and I'm not convinced the results support the claims they're trying to make.

I suggest keeping things simple. Do two separate analyses - one with the reaction time data - and one for participant accuracy. In SPSS, you could conduct a multiple regression for the RT data - though you should also describe the data using a gamma distribution as RT data are usually always positively skewed. You could then conduct a binomial logistic regression for the accuracy data, coding responses as 0 = error, 1 = correct.

I presume the task difficulty manipulation is intended to create variability in participant accuracy. If participants are using a speed-accuracy trade-off strategy, you should observe the following pattern across the two tests: as accuracy decreases, response time also decreases and vice versa.

Blaine Tomkins Article The Improvement of Emotion and Attention Regulation after a ...

Surabhi Lodha If you have repeated-measures factors you can conduct a linear mixed effects regression. In fact, this approach may be better than rANOVA since you will have greater flexibility allowing you to examine nested and crossed effects.