Remember the old saying figures don't lie but liar's figure? Trying to look for simple answers to complex questions leaves people looking for the .4 child since the "average" American home has 2.4 children. LOL
However, mean, median and mode''s are all different "averages" but each tells a different story. Think of a bell curve and standard deviations. Most things fall within "one standard deviation" when you get past 3, you are dealing with extremes. In real estate, the extremes are the luxury market on one end and distressed sales on the other. When you throw out the highs and the lows and deal with the masses in the middle, those numbers tells me the market is improving so the extremes will become less extreme over time.
Another major factor is the accuracy of the sample. In a single zip code if two luxury homes sold one month and non the next, the average "price per square foot" will appear to have tanked because of the small statistical sample.
I have taken many statistics classes, economic classes etc., and have to admit most online valuation models will never be as accurate as an appraiser who personally looks at the home, the neighborhood, the quality, etc., and hand picks comparable s that are in fact comparable. For example, how can anyone effectively write an algorithm to estimate views. How much more is a home worth if it is next to a park verses next to an adult film store? How would an online model know if your home smells of pet odor and gross carpet verses had beautiful hardwood or tile floors?
Remember, these sites are just to give you a rough idea.
Hope this helps!