Adoption of conservation tillage in Australian cropping regions: an application of duration analysis Francis H. D’Emden1, Rick S. Llewellyn2 and Michael P. Burton2 1: Cooperative Research Centre for Australian Weed Management 2: School of Agricultural and Resource Economics, University of Western Australia

Abstract

Agricultural adoption studies in the past have often concentrated on the effects of cross-sectional data on the probability of adoption using techniques such as logit and probit modelling. More recently, Duration Analysis has been used to incorporate time-dependent elements of innovation diffusion by modelling the time to adoption. This paper illustrates the use of Duration Analysis by modelling no-till adoption in southern Australian cropping regions using both cross-sectional and time-dependent variables. The results indicate that some endogenous and exogenous time-dependent variables are significant and highly influential in estimations of the probability of no-till adoption over time.

Most agricultural adoption studies to date do not account for time-varying factors that may influence individuals’ adoption decisions, and concentrate on the probability of adoption at a point in time. Incorporating subjective perceptions relating to an innovation’s value is one way to recognise the dynamic learning process that underpins adoption (Lindner 1987), however, time-varying variables that may help to explain whether and when an individual decides to adopt cannot be included in the typical cross-sectional studies. Treating the process of diffusion and individual adoption decisions as separate entities ignores the fact that the diffusion curve basically represents the growth in adoption over time.

Duration Analysis (DA) is a statistical method that builds on dichotomous choice methods in which individual adoption decisions are modelled using cross-section data, and measurements of aggregate diffusion. Duration Analysis adds a dynamic element to the analysis by combining both individual adoption decisions and the cumulative aspect of innovation diffusion (Burton et al. 2003).
This study uses DA to analyse data on the time to adoption of no-till technology by growers in Australia’s southern grain growing region. The relative importance of a number of personal, locational and economic variables on the likelihood of no-till adoption over a 20 year period is determined. Time-varying variables such as herbicide prices and awareness of local no-till use are included, together with a number of cross sectional variables including perceptions relating to no-till value.

The next section briefly reviews conservation tillage and related adoption literature. This is followed by an overview of the conceptual framework for no-till adoption used in this study. A description of the study area and the variables is followed by an explanation of the DA method and the regression analyses. Implications for no-till and weed management research, development and extension are discussed. The paper concludes by discussing some of the benefits and potential areas of future development for using the DA method in studying the adoption of agricultural innovations.

Adoption of conservation tillage

No-till is an important development in cropping technology as soil surface exposure through full tillage leads to greater soil moisture evaporation and is considered to be a major cause of the decline in soil structure (Chan and Pratley 1998). The benefits of no-till relative to full tillage include:
• improved sowing timeliness
• reduced fuel costs
• improved long- term productivity and water quality
• reduced soil erosion, and
• greater soil moisture retention and water infiltration
(ISTRO, 1997)

No-till sowing systems have been associated with increased cropping intensity and the reduced risk of soil degradation (see McTainsh et al. 2001). However, the removal of tillage as a pressure on weed populations results in a system highly reliant on herbicides. As tillage is reduced, the dependence on chemical weed control tends to increase (Allmaras et al. 1998; Radcliffe 2002; Hooper et al. 2003). In Australia, herbicide resistant weed populations are prevalent in many major grain growing regions (e.g. Llewellyn and Powles 2001), with costlier forms of multiple resistance and glyphosate resistance becoming more common (Powles et al. 1998; Llewellyn et al. 2002; Walsh et al. 2004).

The use of pre-emergent selective herbicides (e.g. trifluralin) are an important component of integrated weed management in no-till systems when combined with a reduction in the use of high resistance risk, post-emergent selective herbicides. It is estimated that the use of pre-emergent selective herbicides in Australian winter broadacre crops has grown from less than 1 million ha in 1990 to nearly 7 million ha in 2003, and that this coincides with the growth in area of Annual ryegrass with resistance to post-emergent selective herbicides (O’Connell, 2004).
There is a wealth of agricultural adoption literature involving conservation tillage, though few studies have addressed no-till adoption in Australia. Most of the existing studies use probit and logit models that rely on cross-sectional data to determine the significance of factors influencing growers’ adoption decisions as measured at a point in time.

It has been commonly observed in conservation tillage adoption studies that larger farms tend to invest more heavily in conservation cropping techniques that involve expensive machinery investments (Rahm and Huffman 1984; Norris and Batie 1987; Gould et al. 1989; Featherstone and Goodwin 1993; Westra and Olson 1997; Wang et al. 2000; Caswell et al. 2001). Explanations for this include the greater capacity for larger farms to make the up-front “lumpy” machinery investment and the greater land area on which to benefit. However, the relationship between changes in farm size over time and adoption of conservation tillage machinery and practices does not appear to have been thoroughly addressed in the past. This may be due to the difficulties in addressing causality, e.g. recognising the possibility that investing in modern reduced tillage technology promotes farm expansion. The DA framework allows for research into the influence of time-varying data on the time to adoption. Changes in farm area were used as a variable in the present study.

Awareness of a soil erosion problem has been found to be a predictor for the adoption of soil conservation measures (e.g. Sinden and King, 1990; Anim, 1999). Wind erosion is a major risk in many of Australia’s southern grain growing regions, particularly in those areas with low (i.e. < 500mm/yr) average annual rainfall (McTainsh et al. 2001). Some of the major benefits of no-till include the reduced risk of soil erosion (Baker et al. 1996) and the favourable influence on seeding timeliness, allowing earlier sowing and crop establishment (i.e. with less opening-season rainfall).

Education and training have also been commonly observed as influential factors in the adoption of conservation tillage practices (Rahm & Huffman 1984; Westra and Olson 1997; Wang et al. 2000; Caswell et al. 2001; Cary et al. 2001). This is typical of a complex innovation that requires system change with relatively high learning demands.

Conceptual framework

In this study it is assumed that growers’ perceptions, along with other cross-sectional and time-dependant farm, farmer and economic characteristics contribute to each individual grower’s subjective utility of no-till as time progresses. The importance of learning, information acquisition and farmer perceptions of innovation specific characteristics in determining adoption has been emphasised in other models (e.g. Lindner et al. 1979; Adesina and Zinnah 1993; Fischer et al. 1996). The study presented here is based on the paradigm that growers hold particular perceptions regarding the effects of no-till, and that these subjective assessments can be important factors in their adoption decision (see Wossink et al. 1997). However, these subjective assessments are under constant review, so the question of adoption probability becomes a question of time to adoption. Conceptually, when the subjective utility of adoption (U_NT) increases relative to conventional tillage (U_CT), adoption becomes more likely, with adoption occurring at the point in time when U_NT> U_CT. So, for an individual grower:
T_NT = f(l,l_t,g,g_t,e_t)
Where:
T_NT = the probability of no-till adoption at time t
l = a vector of cross sectional variables describing the farm characteristics
l_t = a vector of time-dependent variables describing the change in farm characteristics
g = a vector of cross sectional variables describing the grower’s personal characteristics
g_t = a vector of time-dependent variables describing the change in grower’s personal characteristics
e_t = a vector of time-dependent variables describing the change in exogenous economic conditions.

A dichotomous logistic (logit) regression model using maximum-likelihood procedures was initially used to estimate the probability of no-till adoption as at 2003 (see D’Emden et al. 2005), some of the results of which (not shown here) are used as a comparison to the duration model.

Data

This study is based on data from a survey of 384 farmers across southern Australia, comprising 240 no-till adopters and 144 non-adopters. The date of adoption is defined as the first year in which the farmer used no-till for any proportion of their sowing operation.

A phone survey was conducted between March and October 2003 involving growers across a variety of agro-ecological regions within the winter rainfall (400-500mm/yr) dominated grain growing regions of South Australia (SA), Western Australia (WA), western Victoria (Vic) and southern New South Wales (NSW). Phone numbers were randomly selected from publicly available farmer directories. The overall response rate was 51%. See Table 1 for regional sample sizes.

Table 1: Number of responses, district councils and localities comprising survey sample regions

The survey questions were designed to elicit information regarding growers’ tillage practices, along with locational and personal characteristics. Endogenous time-dependent variables were also generated from the survey data (Table 2). Additional exogenous time-dependent covariates are also described in Table 2.

Dependent Variable

No-till has been defined as one pass seeding with narrow/knife points with less than full cut-out (less than 30% soil disturbance) and zero-till as one pass sowing system using discs for minimal soil disturbance (WANTFA, 2004; VNTFA, 2004). In this study, zero-till was assumed to be a subset of no-till, which was defined as using seeding equipment on which only knifepoints or disc openers were used for
soil opening, in one pass with no prior cultivation. The dependent variable used in this model was the time between when the respondent started making management decisions on their farm, to the year in which no-till was first used.

Explanatory variables

Personal characteristics

Table 2 outlines the explanatory variables used in the model. Education, measured by the presence of a tertiary trained (i.e. completed a degree or diploma) contributor to management decisions in the farming business; membership of a local farming group (not including no-till farming groups due to autocorrelation) and extension event attendance, all contributors to accelerated learning, were hypothesised to reduce the time to no-till adoption.

A scale response question format was used to elicit growers’ perceptions of the long-term effects of using a no-till system with stubble retention (NT) relative to a ‘conventional’ system with full tillage and stubble removal (FC). Respondents were given the aforementioned descriptions of the NT and FC systems, and then asked whether they thought long-term use of NT would lead to much lower, slightly lower, the same, slightly higher or much higher levels of 3 different variables (rain needed to allow reliable seeding (RARS), herbicide resistance risk (HR) and effectiveness of pre-emergent herbicides (EPEH)) in comparison to the FC system. These perceived effects of no-till were considered to be potentially important in a grower’s subjective assessment of the relative utility of no-till.

Growers were also asked to state their perceived cost of changing to no-till seeding equipment. It was expected that those growers who thought the costs of changing to no-till seeding equipment to be relatively higher (per unit width) have a longer time to adoption.

Table 2: Definitions of independent variables

A: Does not include no-till farmers associations
B: Average net diesel price paid by Australian farmers, including farm rebates and subsidies.

Farm characteristics

These included variables describing location and average annual rainfall. The state/region dummy variables are relative to Western Australia (WA) as the regions sampled in WA had the highest observed levels of adoption. Time-dependent farm-specific variables included farm size and proportion of farm planted to crop from the time respondents first started making management decisions to the time they adopted no-till, or in the case of non-adoption, 2003. The change in cropping proportion was included in an attempt to capture the change in relative profitability of cropping and livestock enterprises over the study period. It was hypothesised that those with increasingly intensive cropping programs would be more likely to benefit from the no-till cropping technology, and thus have a higher probability of adoption. It is also expected that those whose farm size has been increasing would tend to invest more in conservation cropping techniques involving expensive up-front machinery investments.

Table 3: Descriptive statistics of cross sectional variables by adopter (n = 218) and non-adopter (n = 166) groups.