There are more problems with the calculations - or guestimates as I called them earlier.

Adding a component adds another error source.

So adding an EMI filter or a overvoltage protection component means that the calculation will show a larger probability of failure. But the protection introduced by the new component will probably reduce the probability of the device failing.

This makes it very hard to qualify exactly what the computed MTBF figure actually means or what it actually is worth.

Another thing is that accelerated tests in a climate chamber gives incorrect results. Some components are greatly affected by temperature - for example the retention of information in flash/EEPROM. Some components are much less affected by temperature. So a one-year test at a higher temperature will give a skewed result compared to a 10-year test or a 3-month test, because the amount of acceleration changes the proportions of failures between different components.

The next thing is that much MTBF work gives an estimate of number of units that will fail in one year. So a new HDD have a probability of failure. But the HDD may have a MTBF of 300000 hours which would indicate on average one failure every 34 years. This would indicate maybe a 3 percent probability of a new disk to fail within the first year. But take a number of disks and run them for 34 years, and probably not a single disk will survive even nearly 34 years. The MBTF figures from the manufacturer does not take into account wear and tear - that the bearings can't support so many revolutions of spinning platters, and that the air pressure between heads and platters will result in wear on the surfaces. The MTBF tests conducted by the HDD manufacturer only takes into account mechanical failures caused by production defects - they are the only mechanical fallout that will have time to occur during their inhouse testing. They may be able to get wear-and-tear failures from the head assembly if they continuously run full-stroke head moves but then again, they may not. Many server disks runs their heads continuously 24x7.

So it is always important to sit down and think about the applicability of the MTBF figure. It's so very easy to look at it and think it proves (or indicates) something that it does not. It doesn't even tell if a product with MTBF of 10 years is more reliable than a product with a MTBF of 2 years. The one with 2 years may have seriously good overvoltage protection. The one with 10 years may fail from the first spike on the mains voltage because of bad design.