Biology, images, analysis, design...
Use/Abuse Principles How To Related
"It has long been an axiom of mine that the little things are infinitely the most important" (Sherlock Holmes)




Descriptive surveys (as the name implies) are used to describe the world as it is. Sometimes the only aim of a descriptive survey is to estimate some population parameter, such as the prevalence of a disease, or the density of an insect pest, at one point in time in a single location. This is the situation we have covered in the core text. But often the survey may extend over a period of time (longitudinal survey or monitoring) and/or over space (distribution survey or mapping).

In a descriptive study, usually no specific hypotheses are being tested about factors which may affect the variable of interest. However, associations over time and space between the variable of interest (the response variable) and one or more explanatory variables may be investigated post-hoc. These characteristics are what distinguish a descriptive survey from an analytical survey (where usually a random cross-sectional sample is taken to investigate a hypothesized association between response and explanatory variables) and an observational study (where specific groups are compared to test a hypothesis).

So why look at monitoring and mapping together?

Temporal and spatial data share one very important characteristic when it comes to statistical analysis - their observations tend not to be independent. In other words if you sample an insect pest over time, you will find that samples obtained a short time apart tend to be more similar than samples obtained a long time apart. Similarly the number of cases of a disease tend to be more similar in two nearby locations than in two distant districts. Most statistical techniques assume that your observations are independent, so they are not valid for temporal or spatial data - and may therefore produce highly misleading conclusions when this assumption is ignored.



Longitudinal surveys (monitoring)

Disease monitoring

Epidemiologists may monitor the number of cases of disease in a population (for example cases of Ebola virus in people, or foot and mouth disease in livestock), or they may monitor changes in mortality rates, or changes in birth rates. Another related term, surveillance, is also commonly employed - usually to describe a broader set of activities. These are:

  1. to monitor the number of cases of existing diseases in order to detect any changes in the incidence or prevalence,
  2. to identify the source of infections
  3. to detect new diseases, or new strains of existing diseases,
  4. to detect changes in the systems of livestock management, or movement that may affect the current disease situation, and
  5. to assess the efficacy of control measures, in the event of epidemics.
  6. to obtain data for models, in order to predict the course of an epidemic

Surveillance methods may be passive or active.

  • Passive surveillance relies on routine methods - such as reports of disease cases from local doctors, hospitals, or farmers. Its value depends on what proportion of cases are actually reported, and whether that proportion is constant over time. To improve the sensitivity of passive surveillance, some diseases are notifiable to the authorities by law. 
  • Active surveillance requires sampling the population to find cases. If at all possible, probability sampling should be used so that results can be extrapolated to the target population.

    If disease incidence is very low, and cases may be readily overlooked, an alternative approach is taken. A sentinel population is designated and sampling is restricted to that population. That population may be selected because samples are being taken for another purpose (e.g. pregnancy clinics being used to monitor HIV prevalence) or because the group is exposed to an unusually high risk of infection (e.g. commercial sex workers for HIV infection). In livestock sentinel herds may be introduced in certain high risk areas in order to monitor invasive diseases (e.g. bluetongue) - this is equivalent to the 'canary in coal mine' approach.

    Surveillance is usually carried out by governments as an integral part of disease control programmes. A surveillance system is only effective if the data are gathered, collated, and analysed rapidly - so that up-to-date information can then be made available to those responsible for planning and implementing control measures. This has been made easier by recent advances in data processing techniques.


    Population/environmental monitoring

    Ecologists most commonly monitor certain population characteristics - whether density, pattern of dispersion, age structure or mortality rates. The term 'surveillance' is also used in ecology, again generally referring to a broader set of activities than just monitoring. For example, ecologists may use it to describe long term monitoring and analysis of numbers of migrating insect pests. Unusually high numbers of pests recorded in traps during the growing season are then used to trigger warnings to farmers that control measures may be required.

    Similarly, conservation biologists may establish surveillance programmes to monitor the 'health' of woodlands in terms of their biodiversity.

    For population monitoring, active sampling is usually involved. Again probability sampling is highly desirable, especially at the first stage of sampling.. In the field of wildlife conservation there is often a need to involve local communities in the monitoring process - whether to provide sufficient cheap labour for the job (for example volunteers are frequently recruited for wildlife monitoring in Western countries) or to enhance the feeling of involvement in conservation activities. Such an approach has been termed participatory monitoring.

    Weather conditions, such as rainfall, temperature and humidity, are also commonly monitored - since these may enable you to explain some of the variability in population size, or mortality rate, over time. Often a wider range of environmental variables is monitored - such as food quality and predation pressure. The monitoring data may then be used to investigate associations between population size or age structure and environmental variables as described below.



    Distribution surveys (mapping)

    Other descriptive studies look at distribution of the prevalence or incidence of a disease or the density of an organism over a geographical area, enabling one to map the distribution of the variable. If a fixed network of sampling sites is used, there are advantages to using a systematic grid pattern - especially if you are planning to use a geographic information system to display your data. The map may only show presence/absence of the variable, or it may be quantitative - indicating (for example) number of cases of a disease, or the density of the insect. Some studies describe changes over both space and time - for example, in order to monitor (and perhaps to help predict) the effects of global warming. We look at some (of the many) types of maps below.

    In recent years geographic information systems (GIS) have become very fashionable for mapping purposes. GIS is best defined as an automated system for the capture, storage, retrieval, display and analysis of spatial data. At its heart, a GIS is a map making system - and many users only see it as such. But GIS can also be used as a powerful analytical and predictive tool. Nevertheless, and regardless of how impressive GIS results might appear, their utility unavoidably depends upon the quality of data that are used, the models by which they are interpreted, and the relevance of this approach to solving the problem at hand.



    Investigating associations using temporal or spatial data

    Surveillance and population monitoring data are often used to investigate associations between explanatory and response variables. Explanatory variables may include climatic factors, levels of immunity, or level of control measures. Response variables may include number of cases, incidence rates, population size or mortality rates. You then relate changes over time in the response variable to changes in the explanatory variables.

    Such an approach can be informative, and is often the only way to assess the effects of weather factors on changes in disease incidence or population numbers. But the approach is also fraught with dangers. In particular there is a risk of confounding. Confounding refers to the presence of other variables, which may obscure the effect of the variable you are interested in. Clearly many factors may change over time - not just the ones you have monitored. For example crop losses may decrease over time - concurrent with climatic changes, changes in pest management practices, and changes in pest numbers. Working out what is responsible for what is anything but straightforward. An observed relationship between two variables over time may be purely coincidental because both are related to a third variable. Since many factors are often operating together, mathematical models may be used to increase understanding of the processes at work - or, sometimes, to obscure it!

    There are particular problems if we are looking at only one period of time when two variables happen to start increasing together, as is often the case in epidemiological studies A good example of this problem was when the increased use of the MMR vaccine coincided with an increase in the recorded number of cases of autism. The number of recorded cases of autism increased because of changes in diagnosis. The use of the MMR vaccine increased because it was more effective than the single disease vaccines. Most evidence suggests that occurrence of the two increases, at the same time, was purely coincidental. Indeed some argue that the currently accepted link between bovine spongiform encephalitis (BSE), in cattle, and Creutzfeldt-Jakob disease, in man, may also not be causal. In short, relationships such as this provide very weak inference for a causal relationship.

    If sufficient runs of data are available, monitoring data can be used for forecasting future trends. Forecasting is the prediction of future values that a variable (such as number of disease cases or population size) might take - on the basis of past information about that variable. Forecasting is usually only successful if one has correctly identified the causal relationships between response and explanatory variables.

    topics :

    Remote sensing

    Global positioning system