Excerpt from Open University course TU100 material:
What mistake do you think one might make when investigating messy data like that in Figure 29?
It is difficult to make out any trends or correlations from the data due to the lack of any easily noticeable patterns in the data due to the chaotic nature of the spread of the data points.
Have a look back at Figure 29. How might one select a period of data that would demonstrate that global temperatures are falling?
To select a dataset that correlates with a drop in global temperature by selecting the higher datapoints for the period 1850 – 1910, and the lower data points for the times between 1910 – 1970. Plucking data like that would result in a negative correlation.