nominal, grouping cases into categories among which there is no particular relationship (the continent where a country is located, for example). Others are ordinal, grouping cases into categories that can be ranked higher or lower in terms of some shared attribute. We might, for example, place countries into high, middle, and low GDP-per-capita groups; all in the "high" category would be more affluent than all in the "middle" group, but there would be considerable variation within groups and no consistent variation among groups. Interval-level measurements array cases along a common dimension marked off into units of identical size, but without a point indicating the complete absence of the attribute. The Fahrenheit scale, for example, has an arbitrary zero point: a 1 degree difference is identical across all values, but 40 degrees is not twice as warm as a reading of 20. We might survey several countries asking whether officials are venal or public-spirited, and express the results at the interval level (say, +5 to -5). Such a measure could not, however, tell us a particular country has a total absence of public spirit or that it is twice as venal as another. Finally, ratio-level data also array cases along a dimension marked off in units of identical size, but include a true "zero point." Here, expressions of proportion are appropriate: a country with 50 million residents is twice as populous as its neighbor with 25 million.
Other things being equal, higher levels of precision and measurement are desirable. But there is such a thing as false precision: while it is more useful to know that a country's population density is 255 people per square mile than that it is moderate, it is neither useful nor statistically appropriate to express that measure as 255.348906346 people/mi2. Indeed, one measurement can be more precise, but less accurate, than another: data telling us country X's population density is 255 people/mi2 may be less accurate than an ordinal ranking of "moderate" if the true figure is 75 people/mi2. Level of measurement is an important statistical issue: it is tempting to treat ordinal data as interval-level, for example, but the results can be misleading.