If it's worth saying, but not worth its own post (even in Discussion), then it goes here.
The margin of sampling error is +- 3% while the difference the 1980 percentage and the 2005 percentage is 5%. I do think that a trend which has a p value less than .05 is statistically significant.
I am not sure which data you are looking at.
My link shows the percentage of people who self-identify as Evangelicals. The data starts in 1991 and ends in 2005. The first values (1991-1993) are 41%, 42%, 46%, 44%, 43%, and the last values (2004-2005) are 42%, 39%, 42%, 47%, 40%.
I see no trend.
Your link shows the percentage of people who answer three proxy questions. The data starts in 1976 and ends in 2005. Over that time period one question goes up (47% to 52%), one goes down (38% to 32%) and the third goes up as well (35% to 48%). Do note that the survey ...