Earlier studies have shown that the long-term measure of geomagnetic activity, the aa index, is inhomogeneous and depicts an excessively large (about 12 nT) centennial increase. This has preliminarily been suggested to be due to possible station intercalibration problems in 1957 when the northern aa station was changed from Abinger to Hartland. In the present paper we show that the 3-hourly aa index time series is not uniform but includes systematic jump-like changes in the distribution of the various aa values with each change of stations in 1920, 1926, 1957, and 1980. We estimate how large a change to the aa index was caused by each particular aa value. We find that the changes to the aa index due to different ranges of activity are smooth and fairly similar for all jumps. In 1957 the largest aa values had, at the expense of more moderate aa values, a relatively larger contribution to the jump than in other station changes because the relative station coefficient was somewhat larger in 1957, leading to larger spreading and a higher average level of aa values. However, while this difference could cause a slight overestimate of the aa values, we find that the total changes in the aa index over jumps are in agreement, in both sign and magnitude, with the solar cycle variation. So it is unlikely that the excessive increase of the aa index would be due to erroneously estimated station coefficients.