BACKGROUND: Increasing diagnosis of cancer when the disease is still at early stages is a priority of cancer policy internationally. In England, reducing geographical inequalities in early diagnosis is also a key objective. Stage at diagnosis is not recorded for many patients, which may bias assessments of progress. We evaluate temporal and geographical changes in stage at diagnosis during 2008-2013 for colorectal, non-small cell lung, and ovarian cancers, using multiple imputation to minimise bias from missing data. METHODS: Population-based data from cancer registrations, routes to diagnosis, secondary care, and clinical audits were individually linked. Patient characteristics and recorded stage were summarised. Stage was imputed where missing using auxiliary information (including patient's survival time). Logistic regression was used to estimate temporal and geographical changes in early diagnosis adjusted for case mix using a multilevel model. RESULTS: We analysed 196,511 colorectal, 180,048 non-small cell lung, and 29,076 ovarian cancer patients. We estimate that there were very large increases in the percentage of patients diagnosed at stages I or II between 2008-09 and 2012-13: from 32% to 44% for colorectal cancer, 19% to 25% for non-small cell lung cancer, and 28% to 31% for ovarian cancer. Geographical inequalities reduced for colorectal and ovarian cancer. INTERPRETATION: Multiple imputation is an optimal approach to reduce bias from missing data, but residual bias may be present in these estimates. Increases in early-stage diagnosis coincided with increased diagnosis through the "two week wait" pathway and colorectal screening. Epidemiological analyses from 2013 are needed to evaluate continued progress.