When was climate normal?

It's been a couple years since I took up the question of normal climate, so time for another go.  At that time, I used monthly data from Hadley, and arrived at the observation that if you're younger than 26, you've never seen a month where the global average as as cold as the 1850-2011 average, 317 consecutive months (at that point, now over 330) of warmer than 'normal' temperatures.  I'll cheat and give you some answers first, read on to see how they're established:

• Climate was 'normal' only between 1936-1977
• Every year 1987-present has been warmer than any year before that
• 1976 was warmer than any year before 1926
• 1978 (next coldest year of the recent run) was warmer than any year before 1940

Do read on to see what 'normal' winds up meaning; it's important!  One part of 'normal', as we intuitively think about it, is that you should some times above it, and sometimes below.  Having many consecutive years above 'normal' says that normal isn't really very good.  To help get quantitative about how to proceed, consider this plot of NCDC's data (warmer/colder than the 1880-2012 average -- the length of the entire record).

From this we see that every year 1977-present is above the zero line, much the same as we had in my earlier note.  But, which I paid little attention to at the time, every year in the early part of the record -- everything to 1936 -- is also below the 'normal' (0 anomaly) line.  A long run of negatives is as good a disproof of things being part of the 'normal' as a long run of positives.  Further, 1976 is much colder than all the years which follow and 1937 is much warmer than all the preceding years.  So this makes a good span, 1936-1977, to try to call 'normal'.  It does still satisfy our requirement that to call temperatures climate we want at least 20-30 years, though, at 42, we can't afford to lose many off either end of the record.

This satisfies one of the traits for 'normal' -- no long runs of always above or always below normal at the start or end of the record.  But another trait we often require is no trend.  That is a continuation of the idea that climate is stable in some meaningful sense.  The trend for 1936-1977 is indeed zero.  So for this span, it is indeed reasonable to say that there's a normal climate that weather is bouncing around.  Here's what the data themselves look like, using 1936-1977 for the reference:

This also looks a lot like what intuitions say should be the case for a 'normal' climate, not just the math coming out that way.  As it's 42 years, it's long enough to satisfy our usual requirements for a climate reference.  So now let's look at the whole record against this 'normal':

After seeing this, I have to say that our intuitive demands lead us to a highly unusual 'normal'.  The 1936-1977 span is the only long span with a trend of zero.  The first 30ish years show cooling, next 30ish are warming, the 42 year plateau 1936-1977, and then a 35 (and counting) year warming trend.

If by normal we mean what happens most often, then (for 30 year trends) what's most common is for climate to be changing, not to be some stable reference around which weather bounces.