We used the AgriQuality New Zealand Agribase to select a random sample of farms (geographically stratified by territorial local authority) for the following types of farms: sheep and/or beef, dairy, horticulture (fruit, vegetables, viticulture, nursery, and flowers), cropping (cereal and cropping), and other farms (all other farm types). The following farm types in the Agribase were excluded from the sample: unconfirmed farm type, no farm enterprise or farm type, blank farm type, forestry, woodlot, native bush, tourism, lifestyle, and zoological gardens. Each farm’s predominant farm type for Agribase purposes is selected by the individual farmer concerned. Different farmers may use different criteria for their choice of predominant farm type. Many farms, particularly sheep and beef farms, have a wide range of enterprises other than the one they have selected as the predominant one. This increases the likely variability in the data, because sheep and beef farms do not have only sheep and beef cattle, cropping farms do not have only crops, and so on. However, because it is a feature of farming in New Zealand, we have allowed this source of variability to remain in the sample.

A simple sampling of farms would have resulted in very low numbers of some farm types being sampled. The sample was therefore weighted, so that we would have sufficient farms in each farm type to make reliable estimates of debt levels. Dairy, horticulture, and, in particular, cropping farms were over-sampled, while sheep-beef and "other" farm types under-sampled (Table 1).

Table 1 Number of farms in sample

We subcontracted AgriQuality New Zealand to select the sample from the farms in the Agribase, to make the initial phone call to all farmers selected to obtain their agreement to participate in the survey, to post out the survey forms, and to make follow-up phone calls to encourage the farmers to return the questionnaires (a copy of the questionnaire is in Appendix 10.1). During the recruitment telephone call, further filtering of the sample was done. Farms with no separate farm accounts were excluded, as well as those for which the 1998 farm accounts were not yet available.

The Agribase is a database of farms, not farmers and not farm businesses. Farm debt is held by farm businesses. Unfortunately, no database of farm businesses was available to us; the Agribase was the best alternative. We emphasised to the respondents that they should complete the questionnaire on the basis of a farm business. In most cases, one farm family owns one farm and operates one farm business, and it does not matter which you survey. In more complex cases, though, it can matter. Because the survey was conducted by post, we had little control over the way the respondents interpreted this instruction; however, many telephoned us for advice on this matter.

We were particularly aware of the sensitive nature and confidentiality of the data we were collecting. We therefore took specific steps to protect its confidentiality. A code number was allocated by AgriQuality New Zealand to each questionnaire posted to each farmer. Farmers were asked not to record their names and addresses on the questionnaire. Questionnaires were posted back to AgriQuality New Zealand, who, on receipt of each questionnaire, checked that it had been completed as instructed. AgriQuality New Zealand then forwarded the completed questionnaires to Landcare Research, whose staff entered the data onto a computer database then analysed it using the Excel spreadsheet and SPSS statistical package. At no time was the identification of individual farmers provided by AgriQuality New Zealand to Landcare Research (unless otherwise authorised by individual farmers). Landcare Research then further checked the data and, when any data appeared to be unusual, AgriQuality New Zealand then clarified the data directly with farmers.

In the covering letter posted with the survey form, we gave the names and contact details of the principal researchers, with daytime and evening contact details, so that farmers who required assistance in completing the questionnaire or who had further questions about the survey were able to obtain this assistance. In some cases, farmers’ accountants completed the forms.

The number of farmers eventually contacted was much smaller than the number drawn in the initial sample, because it was difficult to contact many of the farmers. Of those contacted, 227 were outside the sampling frame but still defined as a farm for the purposes of the survey: 193 because the farmers had ceased farming, 31 because the farms were leased out, and 3 because there was a duplication in the list of respondents. These were still defined as a farm because they were likely to be still being farmed, but by someone else. A further 416 were outside the sampling frame but defined as not a farm for the purposes of the survey because they were hobby farms or had no separate farm accounts. Fifteen farms were declared ineligible because they had been operating less than 1 year. A further 554 farmers refused to participate. The main reasons given for refusal were: "too busy", "not interested", and "too personal", but many did not offer a reason. Of the 1478 who agreed to participate, 881 returned usable survey forms, a response rate of 60 percent (Table 2). A total of three reminder telephone calls were made to respondents, if necessary, to encourage them to return their survey forms. One very large dairy farm returned a form showing total liabilities more than 3 times greater than any other farm, and was removed from the sample as an outlier (this farm is not included in the 881 usable returns).

Table 2 Response rate

• Liabilities: loans taken out for farm purchase, the purchase of additional land, for land development, fencing, to erect farm buildings, purchase stock or machinery, for refinancing or to pay for running costs.

• Current liabilities: bank overdraft or short-term loans (of less than 1 year) from dairy company or stock firm etc. They include any items of current liabilities such as unpaid invoices. A new development since the last survey is the revolving credit facility (such as the National Bank’s Freeplan). This type of loan is usually used for seasonal running costs but is also commonly used for the purchase of small capital items such as vehicles and plant. These loans do not have a fixed repayment schedule (as do most term credit arrangements) but they are generally secured by way of mortgage.

• Term liabilities: hire purchase or long-term loans (more than 1 year) from bank or trust company, building society, or insurance company, etc.

• Size of farm operation: We asked farmers to state their gross farm income from all sources. We used this information to classify farms into farm-size quintiles. In the 1991 survey farm size was calculated as the expected value of agricultural output (EVAO) for each farm, based on the number of livestock and size of cropping enterprises.

• Equity: This is the net value of all farm assets after deducting the value of all liabilities. The percentage equity is the net asset value divided by the total value of all farm assets multiplied by 100. To calculate asset values we used the latest government valuation for the land, buildings, and improvements, since the value in the farm accounts is frequently based on historical cost and therefore may not closely reflect market values. For all other capital items we asked farmers to use the book value for assets as stated in their accounts, since we believe this will provide the closest possible consistent and readily available valuation to actual market values.

We divided the farms of each farm type into quintiles, based on the size of their farm operation, measured by income (Table 3). The number of farms in each quintile for cropping, horticulture, and "other" types of farm may be insufficient to place much reliance on the precision of quintile-level data for these three farm types.

Table 3 Number of farms in each farm-size quintile by farm type and total number of farms in survey

In the 1991 study, Pomeroy & Reynolds used the average (or mean) of the data to represent liabilities by farm type. They did not report the spread of the data or the shape of the distribution. Other sources of information about farm debt, with which we compare our results, also presented only the average. For selected data we present the median result as well as the average. This is because the data were skewed by a relatively small number of farms that have high debt levels and this means that the average result may be substantially higher than the median. The median is the figure for which half the sample have lower debt levels and half the sample higher debt levels whereas, with a skewed sample, relatively fewer farms will have debt levels above the average figure.

To determine the representativeness of the sample, we compared some of its demographic characteristics with those of the entire Agribase. First, we compared the percentage of farms of each farm type in the North and South islands. Geographically, the sample is very representative for dairy and horticultural farms (Table 4). It is fairly representative for sheep and beef, and less representative for cropping and "other" farm types. The sample initially drawn was similar in geographic distribution to the Agribase. Almost all departures from geographic representativeness in the final sample were caused by geographic differences in response rates among those in the initial sample.

Table 4 Geographic distribution of farms in Agribase (n=78 123) and survey sample (n=881)

The other demographic data available to us were farm area, stock numbers, and crop area. Because the mean and standard deviation of these values are known for the whole Agribase, we can use a Z test to determine whether the mean values for the sample are significantly different from those of the whole Agribase. We tested for statistical significance at the 5 percent level; in other words, our criterion for a significant difference between the two means was that there was a less than 5 percent probability that the different was not due to chance.

For all farm types the average area of farms in our sample was not significantly different from that of the whole Agribase (Table 5). The only farm types for which there was a substantial difference in average area between the sample and the whole Agribase were sheep and beef and "other" farms. We suspect that farms of these types in the sample were larger on average than those in the whole Agribase because, for these farm types, many small farms were removed from the initial sample because they were lifestyle farms. Median farm areas show a similar trend to average areas. Medians are smaller than averages for both the sample and the Agribase, indicating skewed distributions of farm area.

For dairy, horticulture, and cropping farms, the average number of stock or area of crop for farms in the sample was not significantly different from those in the whole Agribase. For "other" farms, average numbers of beef cattle were significantly higher on sample farms than in the whole Agribase. And, for sheep and beef farms, average numbers of both sheep and beef cattle were significantly higher on sample farms than in the whole Agribase. Again, we suspect that sheep and beef and "other" farms in the sample were larger on average than those in the whole Agribase because many small lifestyle farms of these types were removed from the initial sample. As with farm areas, median farm areas show a similar trend to average areas.

Table 5 Demographic comparisons between Agribase (n=78 123) and survey sample (n=881)