Why are quakes sometimes downgraded?

We collect earthquake data from many different agencies and publish the data as soon as an agency reports a quake. As more data come in, parameters such as magnitude, depth, location in particular can be refined by the agencies, who run extensive numerical calculations which take time to process.
This often results in a downgrade of magnitude - to publish data of a quake fast is necessarily a compromise between data collection and automated processing time versus accuracy, which still requires a manual review by a seismologists; since it is assumed that it is better to overestimate than to underestimate, the first published magnitudes are often (not always) slightly higher than the magnitude that can be calculated when more data is available later.
Typically, the magnitudes and other parameters are well defined only after about an hour.
Various agencies have different methods and different data sets for the same quake; this also can result in slightly different parameters given. We usually prioritize a national data source if available, such as USGS for the U.S., NRCAN for Canada, the British Geological Survey for U.K. and so on.