Daftar isi

1 Data analysis using iSIKHNAS data case studies
2 Concluding remarks
- 2.1 Appendix 1: Extension work using R.
3 References

Data analysis using iSIKHNAS data case studies

Introduction

Pre-requisites

This course is for veterinarians within the Indonesian animal health system. It is assumed that participants will have completed the Excel, Basic Field Epidemiology and Surveillance training modules before this course. If you already know about epidemiology and how to use Excel, then these pre-requisite modules are not required.

Objectives of course

The broader aim of the course is help participants make evidence based animal health policy decisions. This will assist them to improve livestock production and health in Indonesia.

To do this, participants need to be able to access, understand and analyse information on animal health in Indonesia. Fortunately, a new initiative in Indonesia means that Indonesian animal health staff has access to one of the best animal health information systems in the world: iSIKHNAS. This provides staff with large amounts of high quality information (data^[1]) that they can use to make good animal health decisions. Therefore, the objectives of this course are to teach participants to download, understand, evaluate, analyse and interpret iSIKHNAS data.

Learning approach

This course will be taught by analysing real iSIKHNAS data. Three case studies will be presented. During each case study, a question will be asked. Then the question will be answered by participants during practical exercises. Spaces are included after each exercise where you can write your answers. Answers to exercises are provided as Appendix 2. It is generally recommended that you answer the question before reading the answers.

Sometimes, notes on core concepts will be presented before or during case studies to support learning. These are backgrounded with grey to enable you to distinguish these notes from the exercises.

The course is very applied and relevant to Indonesian animal health staff. The three case studies concentrate on: assessment of veterinary services (staff performance), disease management (diarrhoea in cattle) and livestock production (beef self-sufficiency).

The data used in this training course was downloaded in early 2014. This was when iSIKHNAS had been operating for approximately a year in a small part of Indonesia. This early data was used so that we could provide answers to exercises. You may wish to download newer and more complete data to analyse during exercises at the time of your course. Please be aware that if you do, you will not have answers to check your work.

The interim nature of the data means that no real conclusions can be made about the results of data-analyses conducted during this course. Instead, we conducted the analyses and made conclusions to demonstrate and teach data analysis. Over time, more complete data will be available. Then Indonesians will be able to conduct more complete and accurate analyses.

The course is delivered in Excel. Excel was chosen because it is cheap, available to most Indonesian staff and intuitive. Analyses in Excel will allow staff to make some useful conclusions about iSIKHNAS data.

If you intend to do a lot of important statistical work, you will need to learn how to use a complete statistical package instead of Excel. For this reason, we have also included some extension work for those participants who wish to extend their knowledge beyond Excel. This is presented in Appendix 1. Here R, a free online statistical package is introduced. R is one of the most useful software packages in the world. Better still it is free and downloadable from the internet. Appendix 1 repeats case study 1 in R.

Many screenshot videos will be used to assist you in understanding how to do exercises during the course. These can be played on several different software platforms including Windows media player.

Facilitator material

Powerpoint presentation for Day 1

Powerpoint presentation for Day 2

Powerpoint presentation for Day 3

Deciding which test to use presentation

Training schedule

Overview of data analysis

Relevance of data analysis to animal health policy

In order to make good animal health policy, a veterinarian needs to understand the animal health situation where they work. For example, how much disease is present? Or, what is causing disease and how are various interventions working?

To gain this understanding a veterinarian could guess at the situation or they could make assumptions based on their own experience. These are generally poor means of making decisions. Decisions made by guessing are made in the absence of information. Decisions made on their own experience can be useful but are generally based on a very small amount of experience. That is, decisions are based on the experience of only one veterinarian, even if that veterinarian is very experienced.

A better means of decision making for veterinarians is to make decisions based on information that reflects the broader animal health situation. This information can be received in several ways, such as in animal health data, publications, text books and reports. Fortunately, animal health information (data) is now being collected across much of Indonesia. This data is recorded in iSIKHNAS. This data can assist good decision making if it is analysed and interpreted appropriately. The broad objective of this course is to assist you to do this.

Whilst you will learn a lot in the next several days, it is important that soon after completing this course you begin to download and analyse your own iSIKHNAS data. This will ensure that you consolidate your learning, improve your skills and at the same time improve your evidence based decision making. So please, set a day aside next week to do some of your own data analyses using the skills you learn here. Then regularly do some analysis of iSIKHNAS data. Over time your skills will improve.

Introduction to the basic steps of data analysis

There are several recognised steps to analyse and interpret data. These steps are determining an objective for your analyses, data management, describing data and testing hypotheses. Each of these will be briefly introduced here. Then the rest of the manual uses the four steps in the case studies.

Objective

It is important to have a clear and concise objective for your analysis. For example, what is the prevalence of diarrhoea in cattle for 2014? This then allows you to be focused in your efforts and to source appropriate data to address the objective. An objective is then translated into a hypothesis and tested.

Data management

It is important that veterinarians know how to access iSIKHNAS data and use it. We will help you to download iSIKHNAS data. We will also help you to preserve, error check, create and evaluate the iSIKHNAS data.

Description of data

The next step is to describe the data. One purpose of this step is to further check data for errors. Another purpose is to understand the structure and nature of data and the relationships between different parts of the data. In this step, single variable summaries, summaries of relationships between variables and plots are used. This helps you to start hypothesis testing (step 4).

Hypothesis testing

In order to comprehensively address an animal health question it is important to develop a testable question (or hypothesis) from your objective. This hypothesis can then be tested using appropriate statistical tests and you can decide whether the data supports your idea.

Notes: Key concepts for Sampling

Case study 1: Performance measures for veterinary services

Introduction to case study

It is important that a senior veterinarian at local level be able to assess the performance of local veterinary services. A key function of local veterinary services is to investigate livestock disease syndromes that may indicate outbreaks of important diseases.

One means of reviewing performance of a local veterinary service is to see how rapidly important livestock disease syndromes are investigated. We will use this example for case study 1.

The iSIKHNAS data we will use records the occurrence of a number of priority syndromes such as:

Abortion and swollen joints
Biting and behavioural change
Fever in pigs
Limping, drooling and vesicles
Poultry mortality
Sudden death

These are designed to allow early detection of important diseases such as rabies, vesicular diseases such as Foot-and-Mouth disease, highly pathogenic avian influenza, brucellosis and anthrax.

It is important to note that the data used in this training course was downloaded in early 2014. This data was collected when iSIKHNAS had been operating for approximately a year in a limited area of Indonesia. The incomplete nature of our data means that no real conclusions can be made about the results of data-analyses conducted in this course. However, over time more complete data will be available for Indonesians to conduct better analyses.

Skills to be developed during this case study

Develop an objective for a data analysis
Download iSIKHNAS data
Preserve original data
Evaluating data (understand the structure of a dataset, identify errors, understand bias)
Creating new data
Description of data (measures of central tendency, measures of spread, plotting)

List of files for Case Study 1

Data:

isikhnas_priority disease syndromes_March_2013.csv

Videos:

Case study 1_data download video

Case study 1_exercise 3_missing data video

Case study 1_exercise 5_creating new data video

Case study 1_exercise 6_descriptive statistics video

Case study 1_exercise 6b_descriptive statistics without tool pack video

Case study 1_exercise 7_plotting histograms video

Steps in analysis of Case study 1 (exercises)

We will work though the steps of data analysis for this case study.

Step 1: Objective

Before working with the data it is important to have an idea of the question we wish to answer. We call this an objective. Here our general objective is to assess the veterinary services for the speed at which they investigate reports of priority syndromes. We will refine this general objective during an exercise soon (exercise 2).

Open the iSIKHNAS data (isikhnas_priority disease syndromes_March_2013.csv) and examine it. Figure 3 is a screen shot of the priority syndromes worksheet and similar to what you should see. Exercise 1 concentrates on understanding the data. Exercise 2 concentrates on developing a more useful objective. Complete exercise 1 and then exercise 2.

Exercise 1: Question (examine the data)

Break into groups of 2-4 people and conduct the following exercises. Place your answers in the box below this one.

Carefully examine the worksheet columns and the data contained within each of the columns.

What does each of the columns mean?

How many priority syndrome reports are there in the worksheet?

Think about the objective of the analyses we wish to conduct.

Which data columns do you think are particularly useful to address our objective? As a hint, two important columns are highlighted in Figure 3.

Report your findings back to the larger group.

Exercise 1: Answer (examine the data)

What do each of the columns mean?

How many priority syndrome reports are there in the worksheet?

Which data columns do you think are particularly useful to address our objective? As a hint, two important columns are highlighted in Figure 2.

Answers are provided in Appendix 2. Please write your own answers before checking the answers in Appendix 2. This will assist your learning.

Now that you have examined the data and understand it, we need to develop a more specific and useful objective for our analyses. This is the purpose of Exercise 2.

Exercise 2: Question (outline a detailed objective)

Determine a more detailed analysis objective. Hint, what would the difference between the two highlighted date columns in Figure 2 show? Use this to develop a specific objective for the analysis.

Report back to the group with your refined objective.

Exercise 2: Answer (a specific objective)

Write what you think a more specific objective of the analyses should be.

Figure 3: The Priority syndromes worksheet of the iSIKHNAS data download.

Step 2: Data management

In this data management section we will:

Download iSIKHNAS data
Preserve data
Evaluate data
Create new data.

Downloading iSIKHNAS data

This is an important unstructured exercise as it will help you to know how to access iSIKHNAS data for your future work. See the video - Case study 1_data download.avi - for a demonstration of downloading data from iSIKHNAS.

The following steps are how to download iSIKHNAS data.

Go to the website (www.isikhnas.com)
Login with your username (cell number or email) and password
Choose the data set (reports, disease, priority syndrome, then priority disease reports)
Choose the date range you are interested in
Choose the geographic area whose data you wish to examine
Choose run report
Scroll down and click view download and a csv sheet will download
Save this and name it appropriately.

Before continuing the course, pause and practice downloading iSIKHNAS data of relevance to you.

We have downloaded the data you will need for the exercises in this course so we can provide you with exercise answers.

Preserving data

Whenever you download a data file you should save two separate versions of the same file.

One copy should be kept unchanged as an original copy of the data file. You will always have an original and unchanged copy of the data file as it was at the time you downloaded it.

The other copy can then be used as the working file for data analysis. During the analysis process you may make a variety of changes to the data including removing or adding data and variables.

A very good way to do this is to download the original data into a dedicated original data file (for example "Original data" folder in "My Documents"). An alternative practice is to save the original copy with the word _ORIGINAL added to the file name so you can easily identify it. Never work on original data files.

Give each iSIKHNAS download a meaningful name using an appropriate naming convention so you can tell downloads apart. Here we have named our data isikhnas_priority disease syndromes_March_2013.csv.

Having two separate files (working and original) means you can always go back to the original file at any time to start a new analysis or to check what changes have been made in the working copy. If you do go back to the original file to start another analysis, make another working copy of the original file to work on.

See Figure 4 where we have included a screenshot of a possible folder structure with a folder for original data which includes the downloaded Excel file and a separate folder for analysing the data that includes a separate copy of the downloaded data.

Figure 4: Possible folder structure to ensure that original downloaded data is preserved and that analyses occurs in a separate folder with copied data.

Notes: Key concepts about data errors

Evaluating data

Identifying errors and other possible problems in the data

Exercise 3: Question (count missing data entries using Excel)

Count the number of missing data points in the Tanggal diinvestigasi column of the isikhnas_priority disease syndromes_March_2013.csv. Hint: use the filter function in Excel to select only the blank cells (missing data), then count them, or use the countif function in Excel.

Hint: Watch this video for assistance.

Exercise 3: Answer (count missing data entries using Excel)

How many missing values did you count with the filter function?

How many missing values did you count with the "countif" formula?

Exercise 4: Question (selection and information bias).

In your groups, discuss what sort of selection and information bias may be present in the iSIKHNAS data.

Hints: Can you think of any syndromic investigations that may be routinely missing from the dataset? What effect would this have on conclusions based on the data that you do have? This is selection bias.

Can you think of any frequent errors that may occur when you are investigating a syndromic event and making a provisional diagnosis? This is information bias.

Exercise 4: Answer (selection and information bias).

Record some notes from the group discussion on how selection and information bias may occur.

Creating new data It is important to be able to create new columns of data (or variables) using existing columns. This will assist you in doing more detailed analyses of iSIKHNAS data. For example, you will need to do this if you want to determine the time between the report date and the investigation date. In the following exercise you will need to create a new column "jam untuk menyiasat" and populate it with the number hours between reporting and investigation.

Exercise 5: Question (creating new data: hours to investigation)

Create a new column that measure the time between reporting and investigation.

Steps (hint):

Change the format of the columns Tanggal laporan and Tanggal diinvestigasi in the isikhnas_priority disease syndromes_March_2013.csv worksheet to date format (if this is required).

Create a new column heading called time to investigate (masa untuk menyiasat).

Create a formula in the column that subtracts the Tanggal laporan date from the Tanggal diinvestigasi to determine the number of days between dates.

Multiply by 24 to determine the number of hours between reporting and investigation (jam untuk menyiasat).

Convert the answer to a new column of values using paste special (values).

Delete the extra columns, just leaving the jam untuk menyiasat column.

Delete the #NUM! error message using the filter function. These are missing values.

You have now created new data that you can use to investigate the time taken to investigate priority syndromes that allows you to assess the performance of the veterinary services.

Hint: Watch a video for a demonstration

Exercise 5: Answer (creating new data: hours to investigation)

Record in brief note form what you found difficult or easy about exercise 5.

This concludes the step 2: data management. We have considered selection and information bias, we have examined errors and we have created new data. Next, we will describe the data.

Step 3: Description of data

The next stage in analysing this data is to examine the jam untuk menyiasat column and describe it. To do this we will first describe the jam untuk menyiasat column with descriptive statistics and then we will plot the data on a graph. However, before we do this we will provide some notes on descriptive statistics.

Notes: Key concepts for descriptive statistics

Describing a single variable

Exercise 6: Question (calculating measures of central tendency and dispersion of jam untuk menyiasat)

Above we have used 5 simulated values to demonstrate calculations of measures of central tendency and dispersion. Now we will calculate these same measures using real data from iSIKHNAS.

Calculate the mean, median, standard deviation, confidence interval, range and interquartile range for the variable jam untuk menyiasat in the Syndromas prioritas worksheet.

Hint:

Use Excel. Especially use the descriptive analyses tool in Excel.

Hint: See video for a demonstration. This video describes how to conduct the analysis using the Analysis ToolPak in Excel.

If you do not have access to the Analysis ToolPak then you may wish to view the following video on how to calculate the values using individual Excel formulas Watch video now.

Exercise 6: Answer (calculating measures of central tendency and dispersion of jam untuk menyiasat)

Write your answer to each of the parameters in the following table:

Plotting

Now we have finished calculating the measures of central tendency and dispersion on jam untuk menyiasat.

There is an indication that the data is not normal. This is indicated because the mean (750 hours) and median (0 hours) are very different. It is important to look at the distribution of the data in order to be able to choose the most appropriate estimates. For example, will we pay attention to the mean or the median in this case?

To do this we will create a histogram of jam untuk menyiasat. This will give us an idea of the distribution of the data. We will do this in Excel in a moment as an exercise, but prior to this we will present the histogram and discuss it. Here is the histogram of jam untuk menyiasat (Figure 4).

Figure 5: A histogram of hours to investigate a priority syndrome disease report.

By examining the histogram we can see that the jam untuk menyiasat:

Ranges from 0 hours to greater than 3261.6 hours.
There are 43 priority syndromes where investigation was immediate (0 hours). This indicates that a lot of jam untuk menyiasat times were small. This indicates a good veterinary service.
There is a long tail where the time to investigate is long for some reports of a priority disease.
The distribution is not normal. The median and the interquartile range are much better descriptive measures than the mean and standard deviation or confidence interval.

Attention is required to see why some times to investigate are very long. These generally occurred in 2013 when iSIKHNAS was new and may indicate an early data recording problem or that investigation procedures have improved since the recording began. However, further efforts to determine the reasons for long investigation times are required.

Exercise 7: Question (histogram of jam untuk menyiasat in Excel)

Create a histogram of jam untuk menyiasat.

Hint: Watch the video for assistance.

Exercise 7: Answer (histogram)

Place the histogram you produced for the hours to investigate in the space below. You may draw it free hand, cut and paste it from the Excel worksheet into this document electronically or print it and physically paste it in to the space.

Step 4: Statistical hypothesis testing

The fourth and final step in data analyses is to conduct hypothesis testing. However, this is not relevant to this case study where we simply wished to examine the time taken to investigate a priority syndrome report. We will conduct hypothesis testing in the next two case studies.

Summary of case study 1

In this case study you first learnt how to download iSIKHNAS data.

You conducted several important steps in data analysis:

Established an objective.
Managed data (by backing up your original data, examining the data for errors and creating new data).
Described data.

These three common steps should always be undertaken no matter what data you have.

It is important to note that sometimes the way you describe data (step 3) will differ. For example here you used measures of central tendency (e.g. a median) and measure of dispersion (Q1:Q3) to describe the data. This sort of description is only useful for a continuous variable. Other types of data will need different approaches to describe them (see Case study 2 where categorical data is described with contingency tables). Regardless of the type of data, you should always try to describe the data by plotting it on a graph.

In this case study it was not necessary to conduct the 4^th step (hypothesis testing) as we simply wished to explore the time it took to investigate a priority disease report.

This ends case study 1.