In investigations of accuracy, both developmental and cross-sectional in character, most researchers point to the general weakness of the most commonly used measure of accuracy, namely the number of error-free T-units per T-unit. As has already been mentioned, the major drawbacks of this measure were that the frequency of errors within a T-unit was neglected and that errors were not differentiated. On the other hand, measures that looked at the number and gravity of errors often tended to focus more on inaccuracy than accuracy.

These drawbacks also came to light in the present study, where many learners had the same ratio of error-free T-units, but differed in terms of the amount of errors in every unit. For example, in two learners (S12 and S10) the EFT/T was the same (0.38) in the first and the second experiment, respectively. This result could have led to the conclusion that their accuracy was at the same level. Even the length of texts, measured as the number of words (123) and the number of T-units (13), was the same for both writers. However, an analysis of the number of errors in each T-unit showed considerable differences. Student S12 only produced around 0.7 errors per T-unit while S10 made about four times as many errors per T-unit (2.9).

Another sample presents a student with the same rate of error-free T-units in the first two experiments (Figure 6.1). After the first and the second semester this writer's EFT/T was zero, which could be interpreted as meaning that the student did not achieve any progress in accuracy during the first year of learning Swedish. It was only in the third experiment that she began to write with greater accuracy, which represents a clear spurt. However, it is not possible to calculate a reliable growth rate in relation to the previous data collection point due to the above mentioned limitation that we cannot divide by zero. In such cases the researcher would be forced to artificially raise the calculated value up to e.g. 0.01 in order to estimate the growth in accuracy.