Study Design and Logistics - Wetland Techniques - page 30

choosing the initial grid cell to sample, the investigator can then systematically
assign the remaining cells to be sampled. The usual assumption for systematic
sampling is that the study area is relatively homogeneous and thus, the variable of
interest is uniformly distributed across the study area. Occupancy modeling
(MacKenzie et al. 2006 ) frequently utilizes this sampling approach. Advantages
to systematic sampling include being easier to establish sampling units than random
sampling, and it may be more representative (i.e., more precise) because of the
uniform coverage of the entire population (Scheaffer et al. 1990 ; Morrison
et al. 2001 ).

1.10 Errors to Consider in Study Design

A thread linking all aspects of study design is the minimization of errors that impact
results, conclusions, and inference of a study. Because sampling is at the core of any
study design and the primary goal of any study is to produce reliable data, one must
be aware of the potential biases associated with sampling and strive to eliminate or
minimize sources of bias or error. Failure to do so confounds subsequent data
analyses and results, obscuring the true inference and, frequently, contributes to
incorrect conclusions. There are several types of errors that one should be cognizant
of throughout the study design process. Such errors can be categorized as theoreti-
cal, statistical, mechanical or procedural. While investigators need to be aware of
how each type of error affects their study, the best defense against errors dispropor-
tionally affecting one's study is strict adherence to a sound design, sampling
protocols, and data collection.
An example of theoretical error is in the interpretation of statistical results.
Statistical results should be used to support a conclusion or inference based on
totality of evidence from a study, rather than an investigator responding exclusively
to each statistical result. However, errors associated with statistical results can be
found in the inherent uncertainty of statistical tests and expressed in probabilistic
terms. In classical null hypothesis testing, the possibility of conclusion errors should
be considered in the study design. There are two predominate theoretical decision
conclusion errors that can occur in a study. A Type I Error occurs when the null
hypothesis is rejected when it is true. The probability of a Type I Error occurring is

,
which is set by the investigator prior to conducting statistical tests of the data
(conventionally

α

0.05) and commonly referred to as the significance level of a
statistical test (i.e., the probability level at which a test results in a significant
difference between treatments or levels of a treatment). A Type II Error is more
serious than a Type I Error and is defined as the probability (

α ¼

β ) of failing to reject the

null hypothesis when it is indeed false. Determination of β depends on the defined

α -level and the sampling distribution of the estimated variable. More importantly,

one can derive the value of 1 , which is defined as power of the test and defined

as the probability of correctly rejecting a false null hypothesis. Power should only be
calculated prior to conducting a study when computing the required sample size; it

β