1. A number of the problems of sampling farm facts by means of two sample surveys taken in Iowa have been studied statistically. These were essentially problems of sampling efficiency and the detection and measurement of biases and other errors in the data.

2. A sample survey of 800 farms provided estimates for the state of Iowa which were in many cases as accurate or even more accurate than corresponding information provided by the Federal Census, Iowa State Farm Census or Federal Agricultural Marketing Service. On the other hand, it was found that some items cannot be accurately estimated from a small (800 farm) sample survey, but these items are in many cases only of minor importance.

3. The sampling methods used in these experimental surveys were found to be not only relatively free of bias but also satisfactorily efficient. Certain modifications, however, have been recommended. Some principles have been suggested for modifying size of sampling unit for maximum efficiency when certain cost situations are given. Another feature of the sampling method is that it provides a basis for making unbiased estimates of total number of farms and total land in farms in any desired area and, therefore, is independent of any other source of information.

4. It has been found that wide geographical distribution of sampling units (that is, geographical stratification into small areas) Substantially reduces sampling error. Stratification by tenure group would bring only small gains if any.

5. A method has been proposed for determining the best size of sampling unit for given cost situations and for given expenditure levels. It was concluded that the quarter-section grid is an efficient sampling unit under widely varying circumstances. For investigations requiring very short interviews the half-section grid appears to have important advantages. For general inquiries (such as the census, for example), large blocks such as townships appear to be very inefficient sampling units.

6. Matching samples has proved to be an efficient method of measuring differences between years. Compared with samples taken independently each year, matched samples are from 2.5 to over 20 times as efficient, depending upon the item.