I don’t know how many other bloggers post something and then spend the next 24 hours second guessing themselves.  In this case, one of my conclusions yesterday was nagging at me and in a sense of academic honesty I just had to know if in my treatment of the data I had fooled myself and any readers along the way.  So, being the geek that I am I decided to drill down into that one particular survey question to see what else there was to see.

The conclusion that was nagging me was the sensitivity or “high bar” of church attendance as correlated to the growth or decline of denominations.  As part of the analysis I combined some categories in the survey and did not discuss the actual numbers from the survey.  So to remedy that here is an expanded analysis of that single question.  Those who are squeamish over statistics or don’t feel particularly geeky might want to turn away now — this analysis clarifies and qualifies some details but does little to change the overall conclusion I reached yesterday.

To recap, I am working with two data sets.  The first is the National Council of Churches list of the 25 largest denominations, especially the 14 of those that reported growth rates for 2010.  The second is The U.S. Religious Landscape Survey dataset from the Pew Research Center.  The resulting analysis and data manipulation is mine and it
should be kept in mind that “The Pew Research Center and the Pew Forum
on Religion & Public Life bear no responsibility for the analyses or
interpretations of the data presented here.”  For consistency I will again use only the data for the 48 contiguous United States and will not implement their weighting scheme.

In this analysis I want to look at only two questions in the survey.  The first is the multi-part question that established a respondent’s religion or denomination.  This was user supplied and provided some interesting results, as you will see in a minute.  I want to compare that affiliation information against the question “Aside from weddings and funerals, how often do you attend religious services… more than once a week, once a week, once or twice a month, a few times a year, seldom, or never?”

So, for the 14 denominations on the top 25 list that provided information, here are the results for that question.  I have ranked them by growth rate and include total respondents with each answer as well as the percentage.

Well, instead of combining categories I ran correlation statistics on all six meaningful responses.  (You could argue that not responding is meaningful, and looking at the numbers there is a case to be made – why do more Episcopalians not want to respond? – but that is a topic for another time.)   However, from crunching the numbers the first time I noticed that responses from those affiliated with the Catholic Church were frequently outliers, something I pointed out in the first post and something that can be seen in this data set.  It has been observed in other reports that cultural and immigration factors play a larger role in membership numbers for that denomination so I have chosen to exclude those responses from my analysis.

Today, the correlation statistics I calculated include both the linear correlation coefficient as well as the rank correlation.  I won’t go into that latter statistic, except to say that it is a good test for leveraging by extreme values and for none of the responses was that significant, and the only response for which it might have a slight effect is “attend once or twice a month.”

Now it turns out that my combining response categories yesterday may not have been a good way to treat the data because the correlation for “once a week” was not only pretty low, but it was inverse at that.  The only category for which there was a meaningful positive correlation (0.74) was “attend more than once a week.” For “attend once or twice a month” and “attend a few times a year” there are pretty strong negative correlations (-0.84 and -0.81 respectively).  I feel better — While my combining categories may not have been the best move, it appears that it does not substantially change the “high bar” I saw that having the correlation with even “once or twice a month” being related to decline.  At this point I feel I can stick with yesterday’s conclusions.

But having embarked on this data exploration, let me continue with a couple new analyses.

First, using the strongest positive and negative correlations let me ask, “where is the line between growing and declining.”  Now, remember this is only a guideline and not hard and fast, but if we run a linear regression on “more than once a week” we find that using this as a predictor tells us that denominations that have more than 27.5% of affiliated respondents answering in that category were growing.  Looking at the table above (and remembering to skip the Catholic Church) we see that indicator holds up pretty well.  If we do the same with “once or twice a month” we get a predictor that tells us that growing denominations have less than 14.9% of affiliated respondents give that answer.  Again, in the table above this holds up with only one exception.  So while not perfect, these two numbers give a pretty good proxy for predicting growth or decline.

So lets apply these numbers.  First, what about non-denominational churches?  While they don’t represent a denomination, by definition, and we don’t have NCC growth data for them, let’s have a look at the attendance statistics for the three most frequently reported nondenominational categories in the Religious Landscape Survey.

As you can see, all three have “more than once a week” numbers above the indicator, and two out of three have “once or twice a month” numbers below that indicator – and the third misses by only 0.1%.  The indication is that if these were denominations we would expect them to be growing.

OK, lets get close to home — What about Presbyterian Groups?  The survey has 22 self-reported categories of Presbyterians.  Here are a few of the more frequently reported one.

Well, maybe the most important thing about this table is a demonstration of the nature and limitations of surveys.  The first item is the statistics of small numbers.  This dataset works well for the largest denominations, but below the level of the PCA one would like to see a bigger sample.  The second is the self reporting of affiliations and it leaves me wondering if the two different mainline but unspecified categories should be folded into the PC(USA), ignored, or treated as their own group?  And what to do with our liberal and conservative friends?

However, taking the numbers at face value and using the indicators suggested above the only listed Presbyterian branch where we would expect growth is the EPC and the OPC is pretty close.  It is interesting to see the PCA numbers in the same ballpark as the PC(USA).

OK, bottom line — While I need to modify or qualify my attendance calculations from yesterday, the conclusion remains pretty much in tact.  The difference between growing and declining congregations is not in getting Christmas and Easter members to church a couple more times a year (although that would be good) but in fostering an environment were religious faith and participation is taken seriously.

On to the next data set – PC(USA) amendment voting.  Stay tuned.