Math 558 Lecture #1
Stages in Statistically designed experiment
1 Consultation
Copyright By PowCoder代写 加微信 powcoder
2 Statistical design
3 Data collection
4 Data Scrutiny
5 Data Analysis
6 Interpretation of results
Stages in Statistically designed experiment Consultation
Ideally the consultation process should start well before time (about 8 to 12 weeks before the experiment). However, as the consultation process is initiated by the researcher, the statistician usually has little control over it.
If you are contacted very close to the time when the experiment has to start, ask enough question to get clear idea of the experiment (the purpose, resources, timeline).
The questions should be carefully formulated as the researcher/scientist may not be aware of the information you need to design the experiment.
Beware of the perceptions that the scientist/researcher might have about what experimental design is appropriate. There are some handbooks available with limited information on three or four designs that result in a misconception that there are no other designs.
Stages in Statistically designed experiment Consultation
Example 1.1 (Ladybirds)( Bailey, textbook page 2) A company (whose name is not mentioned here for confidentiality reasons) designed an experiment to compare a new pesticide that they had developed, a standard pesticide and no treatment. Their claim was that the new pesticide did not harm ladybirds. Bailey investigated the data and found that the field had been divided into three areas, with one treatment applied to each area. The measurements were then taken on three samples from each area. The researchers who designed the experiment claimed that it was a completely randomized design because ” all samples for each pesticide came from the same area”.They further added that it was completely randomized design because, it was neither a block design, nor a Latin Square and refered to a “respectable” book that gave only these three designs.
Stages in Statistically designed experiment Statistical Design
The (statistical) design of experiments (DOE) is step by step process for planning experiments so that the data obtained from the experiment can be analyzed to yield conclusions. A good experimental design is based on clearly defining the objectives of the experiment and using the available resources in effective manner to draw the required information.
Stages in Statistically designed experiment Data Collection
The data collection process is one of the most important process in statistically designed experiments. We cannot emphasise more on the importance of authentic and credible data. The data collecting procedures should be clear and straightforward. The data should be collected by trained and experienced staff. The data collecting steps should be laid out clearly by the researcher to avoid any mistakes and errors. It is always a good idea for the statistician to discuss in detail the data collection process with the researcher.
Stages in Statistically designed experiment Data collection
An experiment cannot be better than it¡¯s data. Bailey, textbook page 3.
In order to collect the data, you will need a data logger. It can be a piece of paper or a spreadsheet on your electronic device. Make a row for each observational unit and a column for each treatment. Keep some extra columns available to you. You may observe another variable worth recording that was not in the original plan. Tell the researcher not to reorganize the data and to leave them the way they were collected. The reorganizing of the data may result in copying errors. Also, emphasise on the fact that the data collection is too important to be delegated to untrained/ junior staff.
Stages in Statistically designed experiment Data Collection
Example 1.2; Calf feeding ( Bailey, textbook page 2) In a calf-feeding experiment each calf was weighed according to the following plan.
at the time of birthday
On Tuesday nearest to their day of birth The data were collected for eight weeks
Bailey¡¯s comments on the data collected
The data included all these dates , which proved to be mutually in- consistent: some were not Tuesdays and some were wrong length of time apart. When I queried this I was told only the birth dates were reliable: all other dates had been written down at the end of the ex- periment by a temporary worker who was was doing her best to follow the “nearest Tuesday” rule after the event.
Stages in Statistically designed experiment Data Scrutiny
After the data are collected by the researcher, the data files should be sent to the statistician immediately. You need to scrutinize the data carefully for obvious errors, anomalies, outliers or wrong data collection practices as you have seen in the calf feeding experiment. Look for inconsistencies in decimal places, units of measurements and simple descriptive statistics.
Leaf stripe disease: Example 1.3, Bailey, page 3
Each plot was divided into quadrats from which 10 were randomly selected. 1
100 tillers (side shoots) were inspected for the possible infection.
The total number of infected tillers was recorded from each quadrat and the ten numbers were averaged. The official data were based on the average only.
The average for one of the plots was 4.1 while for another was 28. On data scrutiny, it occurred to the statistician that the agronomist forgot to divide the total number by 10.
1each of a number of small areas of habitat, typically of one square meter, selected at random to act as samples for assessing the local distribution of plants or animals. Oxford dictionay
Stages in Statistically designed experiment
Data Scrutiny: Read the examples 1.4, 1.5 and 1.6.
Data Analysis
You data analysis plan should be well thought of before you design your experiment. This means that the information you want to draw from your data affects the way you collect your data. As an example, if you want to study the effect of a new treatment you may decide to use the paired difference t-test and design the experiment to collect the data accordingly. It¡¯s always a good idea to run the analysis on a dummy data set before the real data arrives.
Interpretation
An investigator/scientist generally has little idea of how to interpret the numbers they get from the data analysis. So it will be a part of your job to interpret the results obtained from the analysis in context of the experiment and the research question.
程序代写 CS代考 加微信: powcoder QQ: 1823890830 Email: powcoder@163.com