The origins of reporting bias: Selective but unbiased reporting by early-career researchers?

Doctoral dissertations provide evidence about research practices in early-stage research. We examine reporting bias by manually collecting over 94,000 test statistics from a random sample of German dissertations and their follow-up papers worldwide. We observe selective reporting, as only a fraction of the tests in the dissertations is reported in follow-up papers. Unexpectedly, we find no increase in reporting bias in follow-up papers compared to dissertations nor, generally, reporting bias in dissertations or papers. Self-selection into higher-impact journals based on statistical significance may reconcile our finding of selective yet "unbiased" reporting with prior evidence suggesting pervasive reporting bias.