Communication and the Cancian dip
Adoption of Innovations and the Cancian Dip:
Implications for Communication?
A paper submitted to the Theory and Methods Division of the
Association for Education in Journalism and Mass Communication,
annual convention, Anaheim, CA, August 1996.
ABSTRACT
Since 1967 there has been a debate among agricultural diffusion
researchers concerning the linearity of the relationship between social status
and the likelihood of being an adopter of an innovation. The "Cancian dip"
suggests that when an innovation is at about the 25% diffusion level, members of
the upper-middle class are less likely to be adopters than would be expected by
the linear function. Here, we show the existence of such a dip in the diffusion
of a water quality conservation practice, and explore the role communication
plays in this phenomenon. We find that interpersonal communication among farmers
varies significantly by social status when the dip is present, but does not when
it is absent. Further, these measures of interpersonal communication exhibit the
same upper-middle class dip as does adoption. Strategies for future examination
of the role of communication in an upper-middle class conservatism are
discussed.
INTRODUCTION
The literature on adoption and diffusion of innovations
establishes a critical linkage between communication behavior and successful
adoption programs (Coughenour, 1960; Rogers, 1961). Access to and use of
information channels has been linked to overall diffusion rate (see, for
example, Lin and Burt, 1975; Abd-Ella, Hoiberg and Warren, 1981; Thomas, 1990;
Rogers, 1983, 1995). Different sources and types of information are hypothesized
to be more effective at different stages of the adoption process (see, for
example, Mason, 1964; Thomas, Ladewig and McIntosh, 1990; Blum, 1990; Rogers,
1983, 1995). Mass media, for example, may be more significant as stimuli in
earlier diffusion states, i.e., for early adopters (Rogers, 1985). Media also
appear to be more relied upon during awareness, information and evaluation
stages, while interpersonal communication may be more decisive for trial and
adoption (see, for example, Lionberger and Gwin, 1982; Rogers, 1983, 1995). Here
we examine a contested "special case" situation in the general adoption model,
one which we believe has several implications for better understanding of the
role of communication factors within it.
THE CANCIAN DIP
A lively debate began in 1967 when Frank Cancian first
challenged the accepted wisdom of a linear relationship between social status
and adoption. In diffusion research it was (and still is) held that individuals
with higher social status are more likely to adopt an innovation. While Cancian
did not challenge the overall positive nature of the relationship, he did
present an argument that the relationship between status and class in the
lower-middle and the upper-middle classes deserved special consideration. There
sometimes should, Cancian argues, exist a condition in which the upper-middle
class is less prone to adoption than the lower-middle class, thus creating a
"dip" in the linearity between status and adoption.
Writing in later years, Cancian prefaces a summary of this
proposition in terms of risk and uncertainty. To distinguish between risk and
uncertainty, Cancian follows Knight (1921), whom he summarizes: "In simple
terms, it is risk in situations in which one knows the probabilities of various
possible outcomes of an action; uncertainty in situations in which one cannot
specify the probabilities. (1980, pp. 162-163)" The implication of this
distinction for the diffusion process is direct:
When an innovation is introduced into a community of
farmers from outside, some farmers adopt it immediately, and
some adopt
it in later years. Later adopters usually use the experience of
early
adopters to inform their decision. Thus, uncertainty is greater
for the
earlier adopters than it is for the later adopters. Risk remains
fairly
constant.
This gives us a critical test in the form of two
predictions. First, if uncertainty is meaningfully distinguished
from
risk, poor farmers should adopt more, relative to rich farmers,
in the
early stages of the spread of an innovation. Second, in later
stages,
the rich should be relatively faster adopters.
In the real world, a variety of special
considerations apply to the very rich and the very poor. Thus, I
have
confined my predictions to the behavior of the middle of the
wealth
continuum in agricultural communities. (Cancian, 1980, pp.
167-168).
This, then, is the general basis for Cancian's prediction of a
an upper-middle class conservatism in the early stages of the diffusion of an
innovation: the variable influence of uncertainty and its changing role within
the social structure through time. If uncertainty is an operative mechanism in
the Cancian dip, then it should be reasonable to propose that communication is
also implicated in the phenomenon D and it is toward an investigation of that
possibility that this paper is directed. But it should first be emphasized that
the Cancian dip is far from a generally-accepted proposition. It has, in fact,
generated a fair amount of controversy in an exchange that has now spanned
almost thirty years. We will first, as briefly as possible, review those
exchanges in order to place this present research in perspective for the
uninitiated. Then we will present a discussion and an analysis of the role that
communication may play in upper-middle-class conservatism.
UPPER-MIDDLE-CLASS CONSERVATISM DEBATED
Generally, the relationship between adoption and socioeconomic
status is assumed to be both positive and linear (Rogers, 1995, p. 270).
However, Cancian (1967) proposed that the relationship between economic
stratification and adoption is sometimes non-linear. In his original study he
used a meta-analysis of seven agricultural adoption studies from a range of
locations (Mexico, North Carolina (2), Wisconsin, Iowa Japan, and Kentucky). One
hypothesis tested in the study involved what Cancian termed an upper
middle-class conservatism: the innovativeness (or adoption rate) of those in the
third quartile (upper middle in wealth) will be lower than should be expected by
a linear function. In essence, the innovativeness of the second and third
quartiles are roughly equivalent, or possibly the lower middle class is more
innovative than the upper middle class. This phenomenon is held to occur when an
innovation is in the earlier (about 25%) stages of diffusion in a social system,
where uncertainty with respect to an innovation is more pronounced. Rogers
explains the proposed mechanism and conditions of the Cancian dip:
Cancian's theory says that innovativeness and
socioeconomic status go together at the extremes; that is,
individuals
of highest socioeconomic status are highly innovative, and those
of
lowest socioeconomic status are least innovative. Cancian argues
that b
etween these two extremes, individuals of low-middle
socioeconomic
status are more innovative than individuals of high-middle
status,
especially in the early stages of the diffusion of an innovation
(say
until about 25-percent adoption has occurred in the social
system) when
the degree of uncertainty concerning the innovation is highest.
Later,
when perhaps 50-percent adoption has been reached, Cancian
proposes that
the high-middle individuals catch up and pass the low-middle
individuals, thus resulting in a more linear relationship
between
socioeconomic variables and innovativeness. (Rogers, 1995, p.
270)
Cancian reported support for this proposition in his 1967
publication as well as in another study in 1972. Early replies to Cancian
included projects undertaken first by Gartrell, Wilkening and Presser (1973) and
then by Morrison, Kumar, Rogers and Fliegel (1976). In the former project,
Gartrell and coworkers argued that Cancian's methodology produced a statistical
artifact that itself tended to distort the data in the direction of a non-linear
relationship. Specifically, Cancian cast the dependent variable (innovativeness)
as an ordinal scale and used a binomial model through crosstabulation. This,
Gartrell et al. argued, created a "spurious curvilinearity through measurement
distortion. (1973, p. 401)" Gartrell and his coworkers set innovativeness as an
interval measurement (years since adoption of the innovation, so earlier
adopters have higher values) and utilized least squares tests. After analyzing
data collected on Wisconsin farmers, they concluded that "The relationship
between income and innovation was explained more parsimoniously by a linear
model. (Gartrell, et al., 1973, pp. 407-408)"
The other prompt reply to Cancian, in 1976 by Morrison et al.,
utilized data gathered on Indian farmers. This study also refuted the
proposition of an upper middle-class conservatism. The authors join Gartrell et
al. to point out that Cancian's method of analyzing rank quartiles will
sometimes yield biased results. They argue that real social systems manifest
more of a "triangular" structure, rather than the rectangular structure imposed
by a quartile ranking, and that the distortions caused by imposing quartiles on
a strongly skewed variable leads to this bias. But the researchers went on to
draw a stronger criticism of the basic idea that agricultural innovation is
equivalent to risk:
We have argued for an alternative to Cancian's theory on
the grounds that his operational definition of risk as
agricultural
innovation is inappropriate and that discontinuities in rank
structure
are plausibly related to his findings if his risk measure is
interpreted
as a profit-making rather than an uncertain innovation....
We think innovation is not, as in Cancian's scheme,
the same as risk, even for those who first adopt the innovation.
(Morrison et al., 1973, p. 918)
In a reply attached to Morrison et al., Cancian modified his
quartile approach to a 30/20 split (3:3:2:2) and executed a reanalysis of their
data to defend his theory:
This further analysis of the India data suggests that
the relation of rank and trial is clearly not exponential,
probably not
linear, and possibly not even monotonic. Morrison et al.s'
conclusion
that the India data provides a clear negative replication is not
justified. (Cancian, 1976, p. 922)
In a much larger effort to advance the theory, Cancian published
in 1979 a project incorporating data drawn from some 23 studies providing 49
datasets that presented observations on over 6,000 farmers D all of the original
data was acquired from the previous investigators. This work represented a
fairly powerful reexamination of the question. The results, however, were mixed.
Rogers reflects:
What conclusions were reached from this massive
analysis? In twenty-three of the forty-nine data-sets (each
representing
a farming system in which an agricultural innovation was at
approximately 25-percent adoption), the Cancian dip was
supported in
that the low-middle individuals were more innovative than the
upper
middle. In twenty-six of the forty-nine situations, the Cancian
dip was
not found.
Even though overwhelming evidence in support of the
Cancian dip hypothesis was not found, it is no longer safe to
assume
that socioeconomic status and innovativeness are related in a
linear
fashion, especially at the early stage in the diffusion process.
(Rogers, 1995, p. 272)
The considerable amount of exchange on the concept of an
upper-middle class conservatism produced little of what could be called
absolutely clear evidence, as Rogers notes above, and the matter was certainly
not closed in its first decade. Another series of exchanges was launched by
Frey, Freeman and Lowdermilk in 1979, utilizing data collected in Pakistan.
While they detected a curvilinear relationship between status and adoption, it
was in the second quartile that a dip occurred. The conclude that:
This finding represents a serious anomaly for Cancian's
theory. . . . Consequently, we suggest that Cancian's theory is
incomplete, for it cannot adequately explain the existence of
the
pattern of lower-middle rank conservatism reported here. (Frey,
et al.,
1979, p. 429)
Cancian reanalyzed the Pakistan data and challenged the
refutation. Cancian (1981) stressed the importance of appropriately defining the
community of reference in the process of identifying social classes. He argued
that the Frey et al. project was flawed first in its operationalization of rank
based on cultivable land (versus area owned), and second, in its exclusion of
cases based on missing data and zero values. When Cancian used area owned to
determine rank status and restored missing cases (a majority of which fell into
the lower ranks), he found that "The pattern characteristic of upper middle rank
conservatism is present in the Pakistan data; the anomaly is dispelled.
(1981:640)" In a response attached to Cancian's article, Frey and Freeman (1981)
argue that Cancian was selective in the presentation of his results, failing to
present the several null results found when all possible combinations of
variable definition were applied. They comment:
Regardless of one's interpretation of the two rank
variables, findings do provide partial support for Cancian's
contention
that results vary according to reference community
specification.
However, results do not support the notion that respecification
of
reference community leads to findings that confirm both
hypotheses under
all conditions. (1981:650)
Most recently, two meta-analyses have statistically integrated
the growing body of work on the Cancian dip, and another study has applied more
sophisticated statistical modeling techniques. The first of the meta-analyses,
by Gartrell and Gartrell (1985), draws data from some 34 studies and utilizes
both ANOVA and polynomial regression techniques. While the researchers detected
Cancian's predicted dip, they called the effect too weak to support theory.
"Cancian's theory predicts the existence but not the magnitude of the cubic
effect. Our results support his theory insofar as cubic effects exist; however,
across studies and analyses, the status-innovation relationship appears to be
linear, and Cancian's theory appears to have very small marginal utility in
explaining innovation. (Gartrell and Gartrell, 1985, p. 48)" In a subsequent
article, Lewis joins Gartrell and Gartrell (1989) to reanalyze the data used in
the 1985 article. They compared differences of proportions in the middle ranks
and evaluated Morrison et al.'s hypothesis that a pyramidal conceptualization of
rank will reduce or eliminate any upper middle rank lag. Both of these efforts
lead to a refutation of Cancian's theory.
However, Boyd (1980) confirmed Cancian's hypothesis and also
demonstrated that earlier complaints about introduction of bias through
transformation of interval data into the ordinal level were unfounded. Boyd
utilized an orthogonal transformation of the polynomial terms in a multiple
regression in order to circumvent multicollinearity and to enhance coefficient
comparison and interpretability. His analysis of data from several nations
demonstrated that the cubic model significantly improved prediction when
adoption was at about 25%, but that the linear model was best when adoption was
at about 50%. Interestingly, Boyd's more elegant analysis was not addressed
except in passing by Gartrell and Gartrell (1985) and was not even cited by
Lewis, Gartrell, and Gartrell (1989).
In his 1993 review of diffusion research in rural sociology,
Frederick Fliegel summarizes his observations on Cancian dip:
I doubt that the last word has been said or written
about the linearity or nonlinearity of the relationship between
farmers'
status and their propensity to adopt agricultural innovations. A
tendency toward conservatism among middle-status farmers may be
real but
relatively unimportant, as Gartrell and Gartrell (1985)
conclude. Since
the status-innovation relationship is fundamental to the entire
area of
research, however, it seems likely that the nature of that
relationship
will receive additional research attention in the future.
(1993:92)
Lewis, Gartrell, and Gartrell (1989) and Gartrell and Gartrell
(1985) note that future efforts should begin to focus on the circumstances under
which an upper middle class conservatism might be engendered.
It is also important to recognize that there is
considerable variation in effect sizes among studies. An
important task
for future meta-analysis is the explanation of this variation.
Community-level characteristics . . . and characteristics of the
innovations . . . could be introduced to explain effect
variance. This
would also allow further tests of Cancian's basic postulates,
since the
condition on which they depend could be operationalized.
(Gartrell and
Gartrell, 1985: 49)
However, given the substantial variation in results
which we observe for different communities, future research
should be
directed toward examining factors which may systematically
condition the
effects of status on innovation. (Lewis, et al., 1989:419)
This leads to perhaps the one most important criticism that
might, in retrospect, be leveled at the scholarly exchange on the Cancian dip.
All efforts to date have been directed toward finding, or refuting, this
proposed phenomenon as something that is fundamental to the diffusion process.
To expect that an upper middle class conservatism should always be present when
the overall adoption rate is at about 25 percent is to simultaneously ask far
too much of the theory and to give short shrift to the complexity of the real
world. We think that the considerable body of work on this question points
clearly to the conclusion that sometimes a Cancian dip does manifest itself, and
does so across a range of intensity. This kind of behavior is exactly what we
should expect of a sociometric variable, and effort should be directed D as
suggested above D toward understanding when and how this effect is facilitated.
In the aggregate, upper middle rank conservatism may be inconsequential, but
under some circumstances it may in fact be quite important. The task, then, is
to begin considering what other variables might be implicated when this
phenomenon does manifest itself.
INNOVATION AND COMMUNICATION IN AGRICULTURE
As noted in the introduction, many aspects of communication
should be considered when researching the adoption process. But why,
specifically, should we expect communication to be potentially implicated in a
phenomenon such as the Cancian dip? And if implicated, what role might we expect
communication to play?
Cancian (1980) elaborates on the two underlying principles of
his theory. He points to what he terms the "inhibiting effect of rank," in which
those with higher rank have more to lose. He contrasts this with "the
facilitating effect of wealth," which may bear more closely to our interests
here. Cancian points out:
The facilitating effect of wealth is augmented by the
association of wealth and information and education. Wealthier
people
tend to be better informed. And since the uncertainty that stems
from
lack of information inhibits innovation, it makes sense to
conclude that
better informed people will innovate more in cases where the
innovation
will be to their advantage. (1980:13)
This statement supports the general contention that information,
and thus communication (broadly conceptualized), should present a linear
relationship with innovationDjust as does social status. This linear
relationship has been generally accepted without provision: innovators have a
richer communication profile and are better informed. But the Cancian hypothesis
does indirectly challenge this assumption.
Apparently, only one effort has been made to directly approach
status as a covariate with adoption and any variable resembling communication.
Gartrell and Gartrell (1979) did such an analysis, but looked at the
relationship between awareness and trial at varying levels of status (again
using data from Andhra Pradesh, India). While awareness is not the same as
communication, it might be held that the two are intimately linked. They found,
looking across five separate indicators of status, that as status increased so
did awareness' ability to predict trial. Their analysis looked specifically for
an upper middle class dip in the relationship. They report that:
In short, while there were thresholds in most of these
relationships, there appeared to be little evidence to support
Cancian's
theory of the inhibiting effects of status. When controls were
applied,
the rate at which awareness was translated into trial did not
decline
over the middle or upper-middle ranks of any of the five
dimensions of
status. The most consistent threshold involved the high rate at
which
the elite translated awareness into trial. (p 86)
The researchers did find that above about the 85th percentile of
status, awareness displayed an increasingly greater relationship with trial.
This does suggest a non-linear relationshipDwhich the research did not directly
investigate. The threshold effect reported could be interpreted as evidence of
Cancian's "facilitating effect of wealth."
Considering communication's important role in diffusion, its
close relationship with social status, and the distinct possibility that status
has a more complex relationship with innovation than traditionally assumedDit
makes prima face sense to look to communication in any effort to further
investigate the Cancian dip. Before advancing a set of research questions and
hypotheses, however, it may be best to recapitulate what is known with respect
to communication's role in diffusion within agriculture. Rather than offer a
detailed review of this area of literature, the more salient aspects are
collected in Table 1.
Taken together, these results suggest a rich field from which
hypothetical expectations might be extracted. For example, consider the general
finding that mass communication channels are more important during early stages
of diffusion while interpersonal channels play a more important function in
later stages. This suggests that conditions may change with respect to
communication channels between the 25% and 50% overall diffusion rates, the same
contrast across which the Cancian dip is sometimes observed. Or consider
Gartrell and Gartrell's 1979 examination of status as a covariate of awareness
in its relationship with trial. Higher awareness increasingly promotes trial as
status rises. Might communication variables react similarly as awareness?
Unfortunately, the collection of work on communication offers perhaps an
embarrassment of riches, or at least of variety. Extracting specific hypotheses
for what must essentially be an exploratory investigation could narrow our focus
more than might be desired. Thus, our approach will be to present a nested
series of research questions designed to satisfy the overall query: Is
communication implicated in the Cancian dip?
Given a situation in which we have a variety of communication
variables to evaluate and we: 1) observe an upper-middle class conservatism at
an overall diffusion rate of about 25% in an agricultural social system, and 2)
do not observe a dip when the overall diffusion rate is at about 50% in the same
or a comparable agricultural social system D we can ask the following
hierarchical series of questions.
RQ1: Does communication vary across social status levels?
if so, our attention shifts to differences
in communication when there is a Cancian dip versus
when there is
not, and:
RQ2: Are there aspects of communication that vary by status
when the
Cancian dip is present but not in its absence?
if so, our attention shifts to the
specific function of these communication variables
when the Cancian
dip is present, and:
RQ3: Which communication variables D controlling for status D
significantly increase adoption?
then, with this set of variables, we can
focus specifically on the relationship between
communication and
status within adoption, asking:
RQ4: Which communication variables covary with status in the
prediction of adoption?
From there, we can propose and test a final model as suggested
by the tentative findings. This approach allows us an exploratory framework in
which we can narrow the field of communication variables as we sharpen focus on
the fundamental question at hand. This approach will be developed through the
use of comparative modeling with regression.
DATA COLLECTION METHODS
The overall purpose of this research effort was to investigate
the impact of either USDA Water Quality Demonstration projects begun in 1990.
These are sited in California, Florida, Maryland, Minnesota, Nebraska, North
Carolina, Texas, and Wisconsin. All are in watershed areas with documented water
quality problems related to agriculture. Experts at each site have identified
Best Management Practices (BMPs) capable of reducing agricultural contaminants.
The primary objectives of these projects are to 1) accelerate
voluntary producer adoption of BMPs that will protect surface and groundwaters
from agricultural pollutants, and 2) show how quickly and effectively producers
can modify their current agricultural practices to achieve the first objective.
Producers are being encouraged to voluntarily adopt BMPs compatible with both
at-site water contamination risks and their own production situations. Each
project uses a demonstration farm component and an information and education
campaign to convey appropriate BMP information to producers.
We have attempted to overcome a common restriction with respect
to location and time found in most adoption research (see Lockeretz, 1990): We
examine the above concepts using the same basic measures across (a) multiple
sites nationwide varying in agronomic, watershed, social, and institutional
makeup; (b) several types of water contamination problems and applicable best
management practices; and (c) at least three time points over a five-year
period.
The study used a quasi-experimental research design to compare
producer adoption behaviors within the USDA demonstration areas with producer
adoption behaviors within nearby and matched comparison areas. Large,
representative groups of producers were surveyed in both the demonstration and
comparison areas. Survey respondents were selected using systematic, unaligned
spatial sampling techniques to minimize potential biases. Total response rate
across all sites approximated 70%. Sample sizes differed considerably across
sites due to varying farm populations in the watershed areas, with each sample
size keyed to provide a sampling error margin of +/- 4% or less.
Ideally, one might hope to observe a change in diffusion rate
within one agricultural community that spans the 25% to 50% levels.
Unfortunately, such large scale changes were not observed across the three years
of data collection. Changes were more subtle. However, several comparable
agricultural communities present themselves when the dataset is constricted to a
single time frame and a single recommended agricultural practice. That approach
was used to generate the data subset used in this analysis.
Data were taken from the 1992 sample, which has the added
benefit of pre-dating the more intensive treatment activities carried on across
the span of the project. Thus, samples designated as treatment and control are
less distinct in the 1992 sample and can be combined. A single recommended
agricultural practice is focused on as the innovation: split nitrogen
application. In most of the areas studied in the evaluation project, surface and
groundwater nitrogen loading is a significant environmental problem. In the
split N procedure, the producer divides seasonal use of nitrogen fertilizer into
two or more applications, lessening the high-load runoff that is associated with
single applications. Split N, as an innovation, has the added advantage of being
essentially a production-behavior modification, free of any significant
financial or mechanical burden (for a discussion of conservation practices as
innovations, see Nowak 1984, 1987).
The 1992 baseline dataset was examined for comparable areas in
which adoption of Split N was at about 25% and at about 50%. It was discovered
that samples of field producers (primarily corn and soy beans) in Minnesota and
North Carolina were both near the 25% adoption level. When combined, these
samples yield a working sample size of 217 and an adoption rate of 18.4%.
Samples of field producers in Wisconsin and Nebraska showed adoption levels near
50%. When combined, these samples yield a working sample size of 385 and an
adoption rate of 53.8%. Throughout his work, Cancian (1981) points out that the
ranking of farmers by social status must be done within a local social system.
Thus, measures assigning social rank were extracted from each state dataset
prior to their combinations (differences were minor). Conditions that might be
accountable for the different adoption levels on the various areas are discussed
later in the paper.
Variables to be utilized in the analysis first consist of a
dichotomous indication of adoption as the dependent variable: "Did you use split
nitrogen application on your field crops in the 1991 growing season"? Status was
measured by the total number of acres in operation that year (owned and rented).
Cancian (1979) argues that land in operation is a more appropriate measure than
land owned. Income was also measured, but only categorically. Since income
presents a strong correlation with acres (.49), and acres allow interval level
procedures, acres operated was taken as the status measure. It should be noted
that these data present the skew common to such wealth measures.
Communication behavior, channel exposure, and attitude variables
were incorporated into the instrument. These included questions addressing:
perception of water pollution as a problem, perception of farmers' contribution
to water pollution, informational contact with other farmers, efforts made to
seek and attend to information on water quality, and amount of information heard
from a large variety of sources about water quality protection methods (split N
and others). Specific items appear in the tables below.
RESULTS AND DISCUSSION
Figures 1 and 2 present the analysis demonstrating an
upper-middle class conservatism in the adoption of split N. Confirmation of the
phenomenon was sought through application of both ordinal and interval level
analyses, both of which have been utilized and debated in the literature.
Collapsing data to ordinal level can have the benefit of presenting usefully
simplified pictures, while interval techniques have the advantage of greater
statistical power. We gain added confidence when both present the same result.
For categorical analyses, acres operated were collapsed as
quartile ranks. We used the simpler rectangular ranking approach as opposed to
the triangular ranking scheme also presented in the literature. We do this
primarily because a triangular rank scheme would present a limited number of
cases in the upper rank.
In Figure 1, we see the characteristic pattern that Cancian
describes as an upper-middle class conservatism occurring when the overall
adoption rate is near 25%. Crosstabular analysis shows significant differences
in adoption rates (percent answering yes) between the first quartile and the
second, and between the third and the fourth. The middle classes, however, are
statistically equivalent. A stronger demonstration is found through the
comparison of logistic regression models contrasting linear and cubic equations.
At the 25% adoption level, we see that the linear equation fails to achieve
significance, while the cubic model does. Further, the signs of the coefficients
indicate a positive curve that inflects downward somewhere in the middle range
before again turning upward. Further interpretation of these coefficients is
rather problematic, however. Boyd (1980) presents a methodology for an
orthogonal transformation of polynomials that yields a more useful
interpretation. Our utilization of logistic regression complicates that
approach, probably beyond its benefits, and we prefer not to dwell on the dip
phenomenon per se. From the results presented here, we will conclude that some
departure from linearity, in which some upper segment of the middle range is
depressed, does exist as predicted by Cancian's theory.
This conclusion is bolstered by a comparison with the adoption
rate profile when the rate is near 50%, as shown in Figure 2. The same approach
was taken with these data, which present a clear linear condition. With this
contrast established, we can move to a comparison of the relationship between
communication and status at the 25% and 50% levels.
We first condensed some of the communication variables. The set
of 19 exposure variables were entered into factor analyses (separate analyses
for each adoption rate group). Tables 2 and 3 present these results, which are
interesting and useful. Overall, the results of the factor analysis suggest some
consistency in farmer exposure to various information sources, with logical
groupings presented in both analyses. For comparison purposes contrasting the
25% and 50% groups, we compare (respectively): factors 1 and 2, both involving
government sources; factors 2 and 4, both specialized print sources; factors 3
and 3, interpersonal financial contacts; factors 4 and 1, mixed mediated
sources; factors 5 and 5, closer interpersonal sources. Collapse of these 19
variables into 5 factors greatly simplifies the next phase of the analyses.
The strategy, as developed above, now moves us to a comparison
of the communication variables in the two adoption groups. Here we use a set of
one-way analyses of variance to determine whether communication, exposure, or
environmental attitudes vary across levels of social status (again using the
quartile rankings). This test does not necessarily implicate the communication
variable in the dip phenomenon. Thus, we examine the results for instances where
there is variance in communication across status in the 25% group but not in the
50% group. We use a more selective alpha level of .01, since we are making a
large number of such comparisons.
The results of this phase of the analysis are read from the far
right column of Table 4, where contrasts of interest are indicated in the shaded
areas. This reduces our focus to a set of 8 variables generally addressing
various mediated information exposures, attention, and interpersonal contact. We
will return to Table 4 shortly, but first we address a closer examination of the
8 variables in Table 5. Here we ask which of these 8 variables, those linked to
status when the dip phenomenon is present but not when it is absent, add to a
prediction of adoption behavior. With our focus now exclusively on 25% adoption
group, and our alpha level still at .01, we can see which of the remaining
variables offer statistically significant improvements in predictive power over
the cubic relationship already found between status and adoption.
This analysis is again done through a comparative modeling
approach using logistic regression (note that the Improvement Chi Square
statistic, for the reduction in model deviance, is a functional equivalent to
the F statistic for change in R2 more commonly used in linear regression, see
Lunebord, 1994, chapter 16). We see that only two very closely related
communication variables remain: how often an individual talks to other farmers
about water quality information, and how often that individual is asked for
information by other farmers. Considered separately, the both improve the odds
of a farmer being an adopter by a factor of about 1.5. Another interesting
variable, but one which fails the .01 test, is how much information a farmer
believes he or she has about water quality in general.
Before moving on with the analyses, we direct attention back to
Table 4 with respect to the remaining variables that now have our attention.
Their mean values in the 25% group are shaded in Table 4. Remarkably, we observe
what appears to be a shadow of the upper middle class dip phenomenon. To further
investigate this observation, we apply a post hoc regression of each of these
three communication variables on linear and cubic equations formulated from
status, again comparing model improvement (in R2 here). We find that "having
information" regressed on status yields a significant linear relationship (R2 =
.02, p < .01) but not a significant cubic function (DR2 = .01, p = .07). On
the
other hand, both "contact with other farmers" and "being asked for information"
fail to achieve significance in a linear equation while offering significant
model improvement with cubic terms (for farmer contact, reduced model R2 = .001,
p = .11 and DR2 = .04, p < .01; for being asked, reduced model R2 = .008, p =
.08 and DR2 = .01, p < .05). The correlations between acres and both "talking"
and "asked" are insignificant (both about .08). This, with the missed
significance for "having information" shown in Table 5, brings our focus to
these two remaining, and strongly related, variables (r = .38, p < .01).
The next step in our analysis structure asks if either of these
two communication variables covary with status in the prediction of adoption in
the 25% group. Again using comparison of logistic regression models, an analysis
of covariance allowing interaction was run for both communication variables with
status. Quartile ranking of status was set as three dummy variables (low status
omitted). This analysis of covariance model allows us to test for equal
intercepts and equal slopes. If communication is strongly implicated in the
Cancian dip we might expect to see the communication intercept for the third
quartile significantly lower than the three other groups, and/or the slope for
the third quartile significantly closer to zero than the others.
While both variables do significantly predict adoption ("talk"
Model Chi Square = 9.8, p < .01, "ask" Model Chi Square = 9.6, p < .01), neither
models are improved at the .05 level by either the status dummy block or the
interaction block. We cannot reject the null hypothesis of equal intercepts and
equal slopes. It may be worth noting, however, that the status dummy block does
improve the model in both cases at the alpha = .1 level. This shows, to some
degree, that the variance of means by status seen in Table 3 is preserved in the
relationship that the communication variables share with status as predictors of
adoption. The slopes, however, are not different even at this relaxed level of
confidence.
Thus we may conclude that farmer contact and being asked for
information both increase on average as one rises in social status, but the
relationships that these variables have with adoption is similar across status
levels. If either of these variables were strongly implicated in the Cancian
dip, we might expect to see a different slope for the third quartile, perhaps a
condition in which higher levels of communication are not associated with a
greater likelihood of being an adopter.
This, however, is not reason to abandon our interest in these
communication variables. Their relationship with status is established, and they
both are strong predictors of adoption. From these results, we suggest a final
hierarchical model, which is presented on the bottom of Table 5. Since "talk"
and "ask" are strongly correlated, we are especially interested in their partial
contributions toward predicting adoption, after status is controlled. We would
also be on firm ground asking if they interact. We enter "talk" and "ask" into
the series of equations in that order based on their relative strength as shown
separately in the top section of Table 5. We see that when controlling for one
another and for status, both variables significantly increase the odds of being
an adopter by factors approaching 1.4. Further, we find a significant
interaction that indicates that within the relationships involving "talk" and
"ask," those individuals with highest farmer contact and whom are most often
asked for information have the greatest odds of being adopters.
IMPLICATIONS AND CONCLUSIONS
The Cancian dip does appear to be a phenomenon that occurs under
some circumstances. We do not necessarily believe that this poses a great
challenge to the linear assumption held in so much diffusion research D rather,
it may simply present another strategy worth considering in efforts to improve
diffusion models. If research involving the proposition of an upper-middle class
conservatism can move beyond the standing debate over its existence, and accept
the evidence that it is a situational rather than a universal phenomenon, then
we may begin approaching the task of understanding its causes and consequences.
This present effort has been directed toward that end. It is also the first work
to systematically examine communication within the Cancian framework.
We have observed a variety of relationships that may implicate
communication in the Cancian dip. Although, in a general sense, this work does
not present any significant departure from the accepted knowledge on
communication in adoption, particularly among farmers. Communication behaviors
are generally well correlated with farmer status, as expected from a wealth of
earlier literature. Higher status farmers generally communicate more often, use
more information sources, and so forth. However, some communication behaviors
appear to be significantly more related to status when the Cancian dip occurs
than in its absence. And when we examine those communication behaviors in terms
of their ability to predict adoption, we find interpersonal communication among
farmers and, specifically, farmers asking advice from one another, to be most
strongly operative within the Cancian framework. Finally, it is especially
interesting to observe that those important interpersonal communication
behaviors also present a curvilinear relationship with status D the same
relationship observed between status and adoption.
In a statistical sense, the term "dip" may not be entirely
appropriate, at least for the circumstances of this particular study. It may be
more accurate to conclude that during the early stages of the diffusion of an
innovation into a social system there is relatively less differentiation among
those in the broad middle class both in terms of the likelihood of being an
adopter and in terms of interpersonal communication patterns. Although indirect,
this is some of the strongest evidence offered here that communication may be
causally implicated in the Cancian dip. It is almost as though those in the
upper-middle class are in a temporary doldrums with respect to their potential
to adopt, and those doldrums are either caused by or are reflected in their
interpersonal communication habits.
A nagging question always remains in work with observational
data: what important variables are unmeasured? For our model here, and the
theory from which it is drawn, we might be especially concerned about the
concept of uncertainty. It takes no leap of faith to imagine uncertainty related
to communication. Since uncertainty is at the heart of Cancian's proposition,
this is without question a concept to which we should turn in the further
pursuit of this project.
Interestingly, uncertainty can present a paradoxical condition.
If the third quartile farmers are holding off adoption because of uncertainty,
one expects them to be communicating moreDseeking more information, talking
among themselves, being asked (and asking) questions, and so forth. Yet there
seems to be a holding back, of both communication and adoption, from what we
would expect. Could uncertainty be a barrier, or at least a retardant, to
communication? According to Cancian (1980: 63), we should expect that
uncertainty is higher across the board in the 25% group as compared to the 50%
group. In future work we might look to some measure of uncertainty as a
conditioning variable in an analysis more directly contrasting the two groups we
compare.
On the more practical side, the type of innovation examined
might well be related to the extent of uncertainty. The split application of
nitrogen is a water quality conservation strategy, not a method of improving
production or increasing profitability. In a way, we might expect an enhanced
potential for the Cancian dip when considering the adoption of conservation
methods. For these practices, uncertainty takes on additional dimensions. Not
only must the potential adopter face uncertainty in the economic consequences of
changing an established production practice (especially with little or no
prospect for economic gain), but one must also face some uncertainty in terms of
the actual environmental consequences of the innovation's use. The actual
benefits of some conservation practices are less certain than we might like them
to be.
The degree of uncertainty inherent in a particular innovation
has also been linked to opinion leadership. Rogers (1995, p. 297) suggests that
opinion leaders may be less likely to promote high-uncertainty innovations in
order to better maintain their leadership positions. Opinion leadership is also
likely to be more active in the early diffusion stages: Early adoptors are
traditionally seen as including more opinion leaders in their cohort, in the
sense of individuals who are somewhat more knowledgeable and of somewhat higher
status than the ranks of the early majority who follow. At the 25% diffusion
stage, we would therefore expect to have more opinion leaders in the third
upper-middle quartile, and more "followers" in the second quartile. Thus we
would expect more interpersonal interaction reflecting opinion leadership in the
system at the 25% versus the 50% diffusion stage. And, if the leaders are more
muted due to the higher uncertainty of the innovation, the dip could be a
legitimate reflection of that.
It is also informative to compare the social system in which the
dip is present to the one in which it is not. In this we can observe two things.
First, comparing the differences in adoption rates by quartiles shows that the
largest difference is in the upper-most quartile (about a 50% difference). And,
we see that the upper-middle quartile makes a greater gain than the lower-middle
(about 40% versus 25%) and that the lower-middle is almost identical to the
bottom quartile. Second, we see that when the dip is present, there are
significant differences between status ranks in the amount of communication
activity. These differences are not present when the dip is absent. This may
indicate that in the later stages of diffusion, although communication is more
evenly distributed among social classes, either communication has a greater
effect in the upper classesDor adoption and communication are independent. The
latter seems unlikely, which raises a theoretical point on which we conclude.
Although we have not addressed knowledge gap theory here
(Tichenor, et al., 1970), it may be important to consider when thinking about a
social systemic view of diffusion, communication, and the Cancian dip. Rogers
(1983: 394-410) discusses the knowledge gap in terms of diffusion, especially
with regard to the differential effects of information campaigns. And the
potential for a knowledge gap effect can be clearly related to Cancian's
discussion of the "facilitating effect of wealth" (1980: 13). A differential
ability to utilize new information might take account, at least partially, for
the large jump ahead in adoption that does occur in the upper and upper-middle
ranks, but fails to materialize as strongly in the lower ranks. It is this
differential "rate of acceleration" that appears to erase the Cancian dip as
overall adoption rates increase. Knowledge gap could also provide a useful entry
into a consideration of social-structural factors that might usefully contrast
agricultural systems where dips tend to occur versus those in which they do not,
or a way to consider changing situations within a single social system.
REFERENCES
Aba-Ella, M. M., Hoiberg, E. & Warren, R. D. (1981). Adoption
behavior in
family farm systems: An Iowa study. Rural Sociology 46 (1):
42-61.
Albrecht, D. E. and Ladewig, H. (1985). Adoption of irrigation
technology:
the effects of personal, structural, and environmental
variables. Southern Rural Sociology
3: 26-41.
Andrews, M., Thompson, C., Vuylsteke, P. J., & Berry, M. J.
(1982). Farm
Family Survey. Determining Characteristics and Needs of Farm
Families in East Central
Michigan. E. Lansing, MI: Michigan State Univ.
Anosike, N. and Coughenour, C. M. (1990). The socioeconomic basis
of farm
enterprise diversification decisions. Rural Sociology 47:
234-250.
Blum, A. (1990). Use of Different Information Sources for
Decision Making by
Traditional Farmers in a Progressive Knowledge System.
Research report. Rehovot, Israel:
Hebrew University of Jerusalem.
Boyd, J. P. (1980). Three orthogonal models of adoption of
agricultural
innovation. Rural Sociology 45: 309-324.
Brown, L. (1981). Innovation and Diffusion: A New Perspective.
Methuen, NY.
Bruening, T. H. (1990). Communicating with farmers about
environmental
issues. Paper presented at Agricultural Communicators in
Education research International
Meeting, St. Paul, MN.
Cancian, F. (1967). Stratification and risk-taking: A theory
tested on
agricultural innovation. American Sociological Review 32:
912-927.
Cancian, F. (1976). Reply to Morrison, Kumar, Rogers and Fliegel.
American
Sociological Review 41: 919- 923.
Cancian, F. (1979). The Innovator's Situation: Upper-middle-class
Conservatism in Agricultural Communities. Stanford: Stanford
University Press.
Cancian, F. (1980). Risk and uncertainty in agricultural
decision-making. In
P. F. Barlett (Ed.), Agricultural Decision Making:
Anthropological Contributions to Rural
Development (pp. 161-176). New York : Academic Press.
Cancian, F. (1981). Community of reference in rural
stratification research.
Rural Sociology 46 (4): 626-645.
Coughenour, C. M. (1960). The functioning of farmers'
characteristics in
relation to contact with media and practice adoption. Rural
Sociology 25 (Sept.): 283-97.
Emery, F. E. & Oeser, O. A. (1958). Information, Decision and
Action. New
York: Cambridge University Press.
Fett, J. and Mundy, P. (1990). Disseminating annual crop practice
recommendations via supplements in weekly agricultural
newspapers. Paper presented at the
annual conference of the Agricultural Communicators in
Education, Minneapolis, MN.
Fett, J., Shinners-Gray, T., Schlitz, R., Duffy, K., & Doyle, C.
(1991).
Evaluation of a County Extension Office's Use of Mass Media
and Newsletters: A User
Perspective (Bulletin 45). Madison, WI: Department of
Agricultural Journalism, University
of Wisconsin - Madison.
Ford, S. A. & Babb, E. M. (1989). Farmer sources and uses of
information.
Agribusiness 5 (5): 465-476.
Frey, R. S. & Freeman, D. M. (1981). Stratification and
risk-taking in
Pakistan: A response to Cancian. Rural Sociology 46 (4):
646-651.
Frey, R. S., Freeman, D. M. & Lowdermilk, M. (1979). Cancian's
"upper-middle-class conservatism" thesis: A replication from
Pakistan. Rural Sociology 44
(3): 420-430.
Gartrell, J. & Gartrell, C. D. (1979). Status, knowledge, and
innovation.
Rural Sociology 44 (1): 73-94.
Gartrell, J. & Gartrell, C. D. (1985). Social status and
agricultural
innovation: A meta-analysis. Rural Sociology 50 (1): 38-50.
Gartrell, J. W., Wilkening, E. A. & Presser, H. A. (1973).
Curvilinear and
linear models relating status and innovative behavior: A
reassessment. Rural Sociology 38
(4): 391-411.
Grunig, J, E., Nelson, C. L., Richburg, S. J. & White, T. J.
(1988).
Communication by agricultural publics: internal and external
orientations. Journalism
Quarterly 65: 26-38.
Grunig, J. E. (1971). Communication and the economic
decision-making process
of Colombian peasants. Economic Development and Cultural
Change 19: 580-97.
Hallman, T. (1982, Apr 5). A survey of Georgia farmers: Radio and
television
listening habits. Extension-Editor News.
Jones, C., Sheatsley, P. B., and Stinchcombe, A. L. (1979).
Dakota Farmers
and Ranchers Evaluate Crop and Livestock Surveys. (Report No.
128). Chicago: National
Opinion Research Center.
Jones, E., Batte, M. T., & Schnitkey, G. D. (1989). The impact of
economic
and socioeconomic factors on the demand for information: A
case study of Ohio commercial
farmers. Agribusiness 5 (6): 557-571.
Knight, F. H. (1921). Risk, Uncertainty, and Profit. New York:
Kelley.
Lin, N. & Burt, R. (1975). Differential effects of information
channels in
the process of innovation diffusion. Social Forces 54 (1):
256-274.
Lionberger, H. F. and Gwin, P. H. (1982). Communication
Strategies. Danville,
IL: Interstate Press.
Lockeretz, W. (1990). What have we learned about who conserves
soil? Journal
of Soil and Water Conservation 45 (5): 517-523.
Lunneborg, C. E. (1994). Modeling Experimental and Observational
Data.
Belmont, CA: Duxbury Press.
Mason, R. G. (1964). The use of information sources in the
process of
adoption. Rural Sociology 29: 40-52.
Morrison, D. E., Kumar, K., Rogers, E. M. & Fliegel, F. C.
(1976).
Stratification and risk-taking: A further negative
replication of Cancian's theory. American
Sociological Review 41: 912- 918.
Nowak, P. J. (1984). Adoption and diffusion of soil and water
conservation
practices. In B. English, J. Maetzold, B. Holding & E. Heady
(Eds.), Future Agricultural
Technology and Resource Conservation (pp. 214-37). Ames, IA:
Iowa State University Press.
Nowak, P. J. (1987). The adoption of agricultural conservation
technologies:
Economic and diffusion explanations. Rural Sociology 52 (2):
208-220.
Pampel, F., and van Es, J. (1977). Environmental quality and
issues in
adoption research. Rural Sociology 42: 57-71.
Pounds, D. (1985). Putting extension information where people
will find it.
Journal of Extension 23: 20-23.
Rogers, E. M. (1961). Characteristics of Agricultural Innovators
and Other
Adopter Categories. Wooster, Ohio, Agricultural Experiment
Station Bulletin 882.
Rogers, E. M. (1983, 1995). Diffusion of Innovations. New York:
The Free
Press.
Steele, S. (1979). How many adults use extension? Unpublished
report.
Madison, WI: UW-Extension, University of Wisconsin - Madison.
Thomas, J. K., Ladewig, H. & McIntosh, W. A. (1990). The adoption
of
integrated pest management practices among Texas cotton
growers. Rural Sociology 55 (3):
395-410.
Tichenor, P. J., et al. (1970). Mass media flow and differential
growth in
knowledge. Public Opinion Quarterly 34: 159-170.
Warner, P. D. and Christenson, J. A. (1984). The Cooperative
Extension
Service: A National Assessment. Boulder, CO: Westview Press.
Weaver, J. B. and Miller, B. R. (1982). Information availability,
use, and
economic impact on commercial farms in the Coastal Plain of
Georgia. In College of
Agriculture Experimental Stations, Bulletin 290. University
of Georgia.
Wilkening, E. A. (1956). Role of communicating agents in
technological change
in agriculture. Social Forces 34: 361-367.
Yapa, L. S. and Mayfield, R. C. (1978). Non-adoption of
innovations: evidence
from discriminant analysis. Economic Geography 54: 145-156.
Table 1. Salient Research Results on Agricultural Communication.
RESEARCH FINDINGS
CITATIONS
Adoption is accelerated the more closely farmers are tied to
information channels & networks.
vs.
Economic factors, e.g. financial incentives, are more critical
Rogers, 1983; Nowak, 1987
Abd-Ella et al., 1981; Rogers, 1983; Albrecht and Ladewig; 1985
Pampel and van Es, 1977
Mass communication is relied upon more during awareness,
information and evaluation stages of adoption, while interpersonal communication
is used more for trial and adoption.
depending on:
adoption characteristics of the practice
availability and accessibility of information and channels
utility, practicality and credibility of the information
provided
the appropriateness of timing of the information given the
adoption stage
Lionberger and Gwin, 1982; Rogers, 1983
Nowak, 1987; Thomas, 1990
Mason, 1964; Yapa and Mayfield, 1978; Lionberger and Gwin, 1982;
Rogers, 1983; Grunig et al., 1988; Thomas, 1990
Lionberger and Francis, 1969; Rogers, 1983
Mason, 1964; Rogers, 1983; Blum, 1990
Indications appear that better educated and/or higher
socio-economic strata and/or more organizationally integrated farmers have
access to more information sources, especially more effortful or costly ones.
Communication patterns have also been found to vary by education, experience
level, and size and type of operation.
Wilkening, 1956; Brown, 1981; Nowak, 1987; Anosike and
Coughenour, 1990; Thomas et al., 1990
Integration into local information and assistance networks can
facilitate the adoption process. The availability and characteristics of these
networks, the extent and nature of contacts with representatives of change
agencies, and the position and credibility of these change agents in the local
community can all influence the farmer in the adoption decision.
Nowak, 1987
90% of farmers identified people with special knowledge such as
extension agents, university specialists, and other farmers as important sources
of information.
66% of farm families reported getting production information from
friends, and another 53% from agribusiness dealers and salesmen
Personal conversations with either business contacts or friends
were most important for only 6% of the respondents
Across urban, rural, and rural/urban groups, the most frequently
cited sources of information were professionals or business associates. The
active information seeker first looked to the professional or business sources
and then to the mass media.
40% of Wisconsin farmers believed farm consultants were an
important source of information.
Mass media and "special sources" were extremely important to over
three-fourths of those surveyed, but over 90% of farmers said "people with
special knowledge" were extremely important.
Private consultants to be ranked as the least important place to
get information, followed respectively by farm equipment salesmen, cooperatives,
and television.
Weaver and Miller, 1982
Andrews, Thompson, Vuylsteke, & Berry, 1982
Jones, Sheatsley, & Stinchcombe, 1979
Pounds, 1985
Fett and Mundy, 1990
Weaver & Miller, 1982
Hallman, 1982
Farmers have far more contact with Extension though media than in
person
88% of farm families reported contact with Extension in some form
Steele, 1979; Warner and Christenson, 1984; Fett et al, 1991
Andrews et al., 1982
Differences between farmers information sources depended on their
years of experience, more experienced farmers relied on the Extension service
while younger farmers depended on family and friends. Less experienced farmers
also called on family and friends more for information about credit.
Larger farmers generally considered price information, production
factors, and other information sources more important when making production and
marketing decisions than did smaller farmers.
Ford & Babb, 1989
Weaver & Miller, 1982
Differences in the use of information sources based on farm size
by enterprise: When farmers who raised both livestock and crops were asked
about general information sources such as newspapers, radio, extension and other
farmers, smaller farmers (less than 500 acres) ranked the local newspaper more
useful than larger farm operators did. Larger grain farmers relied more on the
general information sources of national newspaper, television reports and
extension.
Jones, Batte, & Schnitkey, 1989
For buying inputs (feed, fertilizer, or chemicals), farmers used
private firms and cooperatives as the primary source of information. For crop
decisions, however, family and friends were primary, followed by other farmers,
private firms, and Extension. The smaller farmers in this sample used Extension
and farm magazines more than did larger farmers.
Ford and Babb, 1989
For information about broadly defined environmental issues,
farmers rated Extension agents as the most useful of eight information sources.
Local chemical dealers, the Soil Conservation Service, neighbors and friends,
and the Extension specialists were also highly rated. The least useful sources
of information were vocational agriculture teachers and machinery dealers. The
highest rated ways to receive information were through field demonstrations and
county and local meetings. The next most useful were magazines, printed
materials, fairs, and photographs. The least helpful were radio and on-farm
consultation and discussions.
Bruening, 1990
Table 2.
Factor Analysis of Exposure Variables for Overall Adoption Rate
of .25
Wisconsin and Nebraska, n = 385
Information Source Factor 1 Factor 2 Factor 3 Factor 4 Factor
5
Mean SD Government Mediated Interpersonal Mediated Interpersonal
Organizational Specialized Financial General Close
Extension Agent 2.8 0.9 .85
Soil or Water Conservation Agent 2.7 1.0 .78
Extension Publications 3.0 0.9 .67
Farm Meetings, Workshops 2.8 0.9 .60
Demonstrations and Field Days 2.8 0.9 .55
General Farm Magazines 3.6 0.9 .84
Farm Newspapers 3.5 1.0 .81
Specialized Farm Magazines 2.8 1.1 .70
Farm Lenders 2.0 1.0 .75
Landlords/Tenants 2.0 1.1 .73
Independent Consultants 2.5 1.2 .69
Private Newsletters 2.1 1.2 .61
General TV or Radio News 3.1 1.1 .85
TV Farm Programs 2.8 0.9 .67
Radio Farm Programs 2.9 1.1 .66
General Daily or Weekly Newspapers 3.1 1.2 .57
Family Members or Partners 2.9 1.1 .81
Other Farmers 2.9 0.8 .73
Farm Supply Dealers, Suppliers 3.2 0.9 .55
Variance explained 29.5 12.0 9.2 6.4 5.4
Eigevalues 5.6 2.3 1.8 1.2 1.0
How often do you use each of the information sources below?
Responses were on a 5 point scale running from 1 = never to 7 = always. Factors
were determined with an Eigenvalue cutoff of 1.0, principle components analysis
with varimax rotation. Loadings under .5 are blanked. Four factors explain 62.6
percent of total variance.
Table 3.
Factor Analysis of Exposure Variables for Overall Adoption Rate
of .50
Minnesota and N. Carolina, n = 217
Information Source Factor 1 Factor 2 Factor 3 Factor 4 Factor
5
Mean SD Mediated Interpersonal Interpersonal Print Interpersonal
Electronic Government Financial Specialized Close
General TV or Radio News 3.1 1.2 .84
TV Farm Programs 2.9 1.1 .77
General Daily or Weekly Newspapers 1.6 0.9 .67
Radio Farm Programs 2.8 1.1 .64
Extension Agents 3.1 0.9 .88
Soil or Water Conservation Agents 3.0 1.1 .81
Extension Publications 3.3 1.0 .68
Farm Lenders 1.8 0.9 .76
Independent Consultants 1.9 1.1 .67
Landlords/Tenants 1.6 0.9 .63
Private Newsletters 1.8 1.1 .54
General Farm Magazines 3.6 1.0 .77
Specialized Farm Magazines 2.7 1.3 .69
Farm Newspapers 3.4 1.1 .67
Other Farmers 2.8 1.0 .87
Family Members or Partners 2.8 1.2 .77
Farm Supply Dealers, Suppliers 3.2 0.9 .55
Demonstrations or Field Days 2.6 1.1
Farm Meetings or Workshops 2.7 1.1
Variance explained 27.3 13.3 8.9 7.0 5.9
Eigevalues 5.2 2.5 1.7 1.3 1.1
How often do you use each of the information sources below?
Responses were on a 5 point scale running from 1 = never to 7 = always. Factors
were determined with an Eigenvalue cutoff of 1.0, principle components analysis
with varimax rotation. Loadings under .5 are blanked. Four factors explain 62.4
percent of total variance.
Table 4.
ANOVA of Communication Variables by Status Quartiles
Separate comparison of one-way tables at both .25 and
.50 overall adoption rates (.25 n = 385)(.50 n = 217).
All communication variables measured on 1-5 scale with 5
indicating highest, most, or most frequent.
COMMUNICATION OR
.25 RATE
QUARTILE CELL MEANS
ANOVA
INFORMATION QUESTION
.50 RATE
1
2
3
4
F
p
How much have you heard in last 12 months
3.4
3.7
3.7
3.8
4.5
<.01
on what farmers can do about water quality?
3.5
3.6
3.7
3.7
.7
n.s.
How much have you read or heard on WQ from
3.1
3.3
3.3
3.4
.8
n.s.
extension, soil conservation, university?
3.4
3.2
3.3
3.5
.6
n.s.
How much have you read or heard on WQ from
2.2
2.4
2.8
2.9
9.1
<.001
farm supply dealers, consultants, lenders?
2.4
2.4
2.5
2.6
.4
n.s.
How much have you read or heard on WQ from
3.5
3.9
3.8
3.8
3.2
<.05
farm magazines, radio, TV, newsletters?
3.4
3.5
3.9
3.9
4.4
<.01
How much have you read or heard on WQ from
2.8
2.7
2.9
2.8
1.3
n.s.
general news media?
2.8
3.0
2.7
2.6
1.5
n.s.
How much have you read or heard on WQ from
2.6
2.8
3.0
2.9
3.5
<.05
family, friends, other farmers?
3.1
3.0
2.9
2.8
1.2
n.s.
When you come across information on WQ,
3.7
3.8
3.9
4.1
3.8
<.01
how much attention do you pay?
4.0
4.1
3.9
4.0
.2
n.s.
Overall, to what extent do you need more
3.4
3.2
3.3
3.4
.8
n.s.
information on what you can do about WQ?
3.6
3.5
3.4
3.3
1.0
n.s.
To what extent have you looked for information
2.8
2.9
2.9
3.1
1.3
n.s.
on how to protect/improve water quality?
3.1
2.9
2.9
3.2
.8
n.s.
How often do you talk to other farmers about
2.5
2.8
2.8
3.1
6.4
<.001
what you could do to protect water quality?
2.8
2.6
2.7
2.9
1.0
n.s.
Compared to other farmers in your area, how
2.7
3.1
2.8
3.3
7.1
<.001
much information do you have on WQ?
2.9
2.9
3.0
3.4
3.0
<.05
Compared to other farmers in your area, how
2.5
2.9
2.7
3.0
5.2
<.01
likely are you to be asked for info. on WQ?
2.6
2.6
2.7
3.3
3.7
<.05
Exp. Factor 1: Government, Organizational
-.07
.12
-.03
-.01
.6
n.s.
Exp. Factor 2: Interpersonal Government
.00
.00
-.04
.01
.1
n.s.
Exp. Factor 2: Mediated Specialized
-.19
.3
.03
-.13
4.94
<.01
Exp. Factor 4: Print Specialized
-.29
-.04
.2
.23
3.3
<.05
Exp. Factor 3: Interpersonal Financial
-.67
-.16
.28
.53
30.7
<.001
Exp. Factor 3: Interpersonal Financial
-.3
-.01
-.14
.47
6.3
<.001
Exp. Factor 4: Mediated General
.14
-.1
-.02
-.02
.9
n.s.
Exp. Factor 1: Mediated Electronic
.1
.07
-.07
-.02
1.0
n.s.
Exp. Factor 5: Family, Farmers, Dealers
-.27
-.13
.21
.21
5.3
<.01
Exp. Factor 5: Family, Farmers, Dealers
.3
.05
.12
.15
2.6
<.05
In your opinion, to what extent is water
2.8
2.9
2.8
2.8
.9
n.s.
pollution a problem? *
3.1
2.9
2.9
2.9
1.5
n.s.
How much impact do farmers like yourself
2.8
2.8
2.9
2.8
.3
n.s.
have on water pollution problems? **
2.9
2.8
2.7
2.9
.9
n.s.
* Index of questions about water pollution at national,
state, county, and local level, alpha = .81
** Index of questions about farmer impact at national,
state, county, and local level, alpha = .82
Table 5.
Testing Communication Variables and Final Model
for overall adoption rate of .25 in Wisconsin and
Nebraska (n = 385)
Coefficients and odds are from full models. Dc2
indicates model improvement made by inclusion of the variable in reduced cubic
equation: log(ADOPT) = -2 + 0.001(ACRES) - 2.2-7 (ACRES)2 + 8.7-12 (ACRES)3
where reduced model c2 = 8.9 with d.f. = 3 (p =
.031). Probability values are for Dc2 where d.f. = 1.
VARIABLE DESCRIPTION
B odds
Dc2
p
How much heard on what farmers can do to improve
water quality?
.27 1.31
2.5
n.s.
How much read or heard about WQ from supply
dealers, consultants, lenders?
.08 1.08
0.4
n.s.
When you come across information on WQ, how much
attention do you pay?
.21 1.23
1.3
n.s.
Exposure Factor 2: Mediated Specialized
.21 1.23
2.1
n.s.
Exposure Factor 5: Family, Farmers, Dealers
.09 1.09
0.4
n.s.
Compared to other farmers, how much information do
you have on WQ?
.30 1.35
4.3
.038
Compared to other farmers, how likely are you to
be asked for info. on WQ?
(ASK)
.39 1.48
8.1
.004
How often do you talk to other farmers about what
you could do to protect WQ?
(TALK)
.43 1.53
7.1
.008
FINAL HIERARCHICAL MODEL:
log(ADOPT) =
- 2.1 + 0.001(ACRES) - 2.3-7 (ACRES)2 + 8.4-12
(ACRES)3
Model c2 = 10.1 ( p = .018)
- 3.1 + 0.001(ACRES) + 3.7-8 (ACRES)2 - 1.7-11
(ACRES)3 + 0.42(TALK)
Improvement c2 = 6.9 ( p = .008)
- 3.7 + 0.001(ACRES) - 1.6-7 (ACRES)2 + 4.3-12
(ACRES)3 + 0.3(TALK) + 0.3(ASK)
Improvement c2 = 4.1 ( p = .043) odds = 1.35
1.36
- 0.9 + 0.001(ACRES) - 2.1-7 (ACRES)2 + 7.9-12
(ACRES)3 - 0.7(TALK) - 0.7(ASK) + .34(TALK*ASK)
Improvement c2 = 5.8 ( p = .016) odds = 0.49
0.51 1.40
|
