Simple answer. Alexa is junk. Do not even look at it. Only for the most active sites will the Alexa number mean anything. Otherwise, it will be a misleading metric that should not be paid attention to.
The reason is that Alexa has to extrapolate their results based upon a self-selective group. Alexa says that it's toolbar users are diverse, however, clearly it is a self-selected group regardless - meaning that the data sample is provided by only those who would install the Alexa toolbar. In the statistics game, that is a disaster. Only random samples work. Even then, the larger the sample size the better. Think about it. Who installs the Alexa toolbar anymore? Especially when they get a new computer? The Alexa audience is becoming narrower as fewer people install the toolbar and those who do fit within a smaller group of those who would. It was a flawed system right from the start. At least from a statistics analysis point of view.
To that end, sites with less traffic have less sample data to make any clear indication. This is why Alexa varies so widely and hence why you should not pay attention to the metric at all.
[Update]
I had taken the time to definitively prove why Alexa metrics are junk on my own website found here http://www.closetnoc.org/?i=definitive-proof-that-alexa-sucks (I hesitate to link to my own site, but this is well worth a read for a complete understanding). Here are just some of the takeaways that I can add here without long explanation:
- The Alexa Toolbar is only compatible with 2/3rds of the popular web browsers.
- The Alexa Toolbar is largely incompatible with mobile search which is at least 1/3rd of search.
- Alexa cannot know about the statistical population of users who use browsers that are not supported by Alexa, mobile users, developing countries, non-Alexa users, and so on.
- In statistical analysis, it is impossible to correct for biases using a population you do not know.
- The Alexa Toolbar users are self-selective toward blog users, Twitter users, Facebook users, and other social media.
- The Alexa Toolbar users are self-selective toward content and sites that are trend based and not research based.
- The Alexa Toolbar users are self-selective away from research sites and search engine usage.
- The Alexa Toolbar users trend toward popular topics/sites and those who concern themselves toward popular topics- this means a younger demographic and a demographic that does not include older or senior users.
- The Alexa Toolbar not being mobile compatible, has lost many millions of young users in the past two years and growing- trending toward a smaller audience that is becoming more middle aged.
- The Alexa Toolbar is losing popularity and new computer users, who once used the Alexa Toolbar, are opting not to re-install the toolbar on a new computers.
- The Alexa bug is not 100% accurate with complaints on it's effectiveness to capture accurate usage.
- The Alexa bug cannot compensate for a population that is not known.
- Alexa favors Alexa ratings and reviews with bias.
- Alexa favors trend sites with bias.
- Alexa favors social media usage with bias.