It's like anecdote hell in here. Your individual experience doesn't dictate overall reliability. You are a single point of data. Reliability is measured as a probability, in which case a dataset of 1 is useless.
The good news is that automobile reliability is measured in terms of defects per thousand, and cars have never been more reliable than they are today . We've become exceedingly good at precision manufacturing.
On to the actual survey. Keep in mind what CarMD does/is. Their product is a code reader that can be used to diagnose automobile issues. Their claim is that the report is "unbiased", but I'd say that's a misnomer. It may be free from the bias associated with customer surveys, but the data itself may be biased by product itself. That is to say, the nature of the product is going to dictate their data set.
I can't see why a new car buyer would purchase a CarMD unit or subscription when service at the dealership is free, so I expect that their data set is from older cars. That still makes it useful, but I wish they were more clear about what their data set is composed of. What is the age distribution (min/max/avg) of the cars in the database? A histogram of the age by year would be nice. Also, what is the distribution by make? Another histogram of the records by make would be nice.
The link says the data is available somewhere. If I can get my hands on it, I'll do some looking in to these questions and post back here. One thing is certain: this is actual data collected from the field, so we shouldn't be so dismissive. It may not answer the question we're asking, but it certainly contains answers to some questions.
1 - http://www.cnbc.com/id/46397659/Car_...ts_Record_High