Data Science Centre (DSC) Data Preparation Policy: Guidelines for managing and preparing your data for statistical analysis

Transcription

1 Data Science Centre (DSC) Data Preparation Policy: Guidelines for managing and preparing your data for statistical analysis Dr Fiona Boland, RCSI Data Science Centre & HRB Centre for Primary Care Research Version: 1.0 Date: October 2017 Table of Contents 1. Sharing your data with the DSC Introduction Choosing a file format Organising and entering data Codebook / Data dictionary Backups Checking your data data cleaning Data protection Useful resources Appendix A Saving your Excel spreadsheet as a CSV file ~ 1 ~

2 1. Sharing your data with the DSC Providing the necessary documents and ensuring data is in an organised manner will reduce the likelihood of errors and facilitate the most efficient and timely analysis. The following files should be provided to the DSC: 1. A brief description of the study design and research questions. 2. A structured and organised raw data file in an appropriate file format ensure it is structured as outlined in section 4 below. 3. A code book describing each variable and possible values see section 5 below. Please note that the DSC will return any data file(s) not adhering to the standards set out in this policy for further formatting and/or cleaning. 2. Introduction An important aspect of any project is recording, organising and preparing your data for statistical analysis. When carefully planned and implemented from the beginning of a project it can save you a lot of time and hassle. Furthermore, often data may need to be shared with a colleague or the Data Science Centre (DSC). If the DSC is receiving the data, they should be able to work with it in the format in which they receive it. This will speed up the amount of time required for analysis and obtaining results. The aim of these guidelines is to provide researchers with information and advice on how to manage and prepare research data for statistical analysis and / or sharing. 3. Choosing a file format In planning a research project, it is important that you consider which file formats you will use to store and prepare your data for statistical analysis. You should consider what formats will be easiest to share with the DSC and /or colleagues for future projects. You can enter your data directly into a statistical package or you can use a spreadsheet (such as Excel). Entering data directly into a statistical package is preferable, you can label the variables within your data file (variable labels) and label the values for your variables (value labels). ~ 2 ~

3 Variable Labels: Variable labels are composed of a few words that describe what a variable represents. If the variable labels are properly formatted, they will show in output tables and graphs, instead of variable names. Value Labels: Value labels are labels for coded variables in a data set. For example, "Gender" may be coded 0 (Males) and 1 (Females). Excel Please save your Excel spreadsheet as a comma separated values' or CSV file. See Appendix A for saving your Excel spreadsheet as a CSV file. SPSS and Stata You can also use SPSS or Stata to enter your data. This allows you to specify missing values, attach labels to numeric values and several other useful features to speed up data entry. You can send the DSC the SPSS file (.sav) or the Stata file (.dta). R You can send the DSC the R data file (.Rdata). Other statistical packages Data can be read from just about any statistical package, but please check with the DSC beforehand. 4. Organising and entering data To ensure that your data can be read by a statistical package, and analysed promptly, you need to follow a couple of simple rules: 1. Just one spreadsheet: Keep all your data in one spreadsheet; do not use multiple spreadsheets. Do not use different spreadsheets for different groups of participants or different time points. For example, if you are collecting data at three separate time points, do not use a worksheet for each time point. ~ 3 ~

4 Bad Example: Here there is an excel file with a spreadsheet for each time point. Good Example: There is only one spreadsheet where the variables Weight_1, Weight_2 and Weight_3 correspond to weight at time 1, 2 and 3 respectively. ~ 4 ~

5 2. Be consistent: Enter your data in a consistent manner (variable names, labels, spelling and codes for variables, dates, etc.). Please note that capitalisation is acceptable as long as it is used consistently throughout the data file. If in doubt please use lower case only. Examples: a. If you are collecting data at various time points, for example weight over ten weeks, and you begin labelling using weight_wk1 for week 1 and weight_wk2 for week 2, continue in this manner. Do not change to week3_weight or wk10 weight. b. Gender: If you pick Male and Female, do not occasionally use M or F. c. Dates: Pick a format and stick to it throughout the spreadsheet. The main thing is to be consistent but if in doubt use dd/mm/yyyy, where dd is a 2 digit day, mm is a 2 digit month and yyyy is a 4 digit year. d. Subject identifiers: If you begin by using PT001 for patient 1, use this format throughout. Bad Example: Good Example: Inconsistent date formats Inconsistent ID numbers Inconsistent coding for Sex Consistent formats for all three variables; ID, DOB and Sex. ID DOB Sex PT001 31/12/1966 f PT Jun-82 m 3 17/03/1977 Female PT004 10/11/1972 F 5 27/04/1960 male 6 23/05/1973 female 7 13/06/1974 M PT008 05/07/1962 M 9 20-Sep-84 Male PT010 08/10/1989 female ID DOB Sex PT001 31/12/1966 female PT002 01/06/1982 male PT003 17/03/1977 female PT004 10/11/1972 female PT005 27/04/1960 male PT006 23/05/1973 female PT007 13/06/1974 male PT008 05/07/1962 male PT009 20/09/1984 male PT010 08/10/1989 female ~ 5 ~

6 3. Row 1 = variable names: The first row of the spreadsheet should always contain the names of the variables. Make sure that each variable has a unique and informative name. Try not to use a name that is very long. Keep the names of the variables in the first row only - do not use the second row as some sort of continuation. The second row of the spreadsheet should contain data on the first case. As a general rule, variable names should not begin with a number, contain spaces or contain arithmetic symbols like +, >, %,? or!. Use underscores in place of spaces or /. For example, if you have several time points you could use walk24, walk48, walk72 or weight_1, weight_7 etc. Be careful not to leave spaces at the beginning or end of variable names, for example walk24 is different to walk24 Bad Example: 1. Variable names and information about the variables spans the first 3 rows. 2. Date of birth and Usual Activities contain spaces and Self/care contains a /. Section 1 EQ5D EQ5D EQ5D ID Date of Birth Sex Mobility Self/care Usual Activities /12/1966 female /06/1982 male /03/1977 female /11/1972 female /04/1960 male /05/1973 female /06/1974 male /07/1962 male /09/1984 male /10/1989 female ~ 6 ~

7 Good Example: ID DOB Sex EQ5D_Mobility EQ5D_Selfcare EQ5D_UsualActivities /12/1966 female /06/1982 male /03/1977 female /11/1972 female /04/1960 male /05/1973 female /06/1974 male /07/1962 male /09/1984 male /10/1989 female One row = one case: Each row of the spreadsheet should contain all the data from one case (usually a person or animal). Have a unique ID number for each case, to cross-link paper and computer records and to check for duplicates or omissions. 5. One column = one variable: A variable contains one item of information only. a. Do not put things like '120/80' into a spreadsheet. 120/80 is two things, systolic and diastolic pressure. They must be recorded in separate columns. Bad Example: ID Sex Age Systolic/Diastolic / / /79 Good Example: ID Sex Age Systolic Diastolic ~ 7 ~

8 b. If you have an item/question where multiple responses can be ticked (e.g. survey), put each response into a separate column, coded yes/no or 1/0 (as opposed to one column listing the options people selected). Generally, 0 (zero) is used for absent and 1 for present. Bad Example: In a survey Q1 has 5 possible options that can be selected and the question states to select all that apply. PT001 selected options 1 and 2 for Q1 and PT002 selected options 2 and 4. Recording the data in the following manner makes it impossible to analyse the data. ID Sex Age Q1 PT ,2 PT ,4 PT ,3,4 Good Example: In the following table a column is assigned to each option in Q1 (i.e. Q1_1 for option 1, Q1_2 for option 2 and so on.) If PT001 selected options 1 and 2, then Q1_1 and Q1_2 are denoted with a 1. PT001 did not select options 3 or 4; hence, Q1_3 and Q1_4 are denoted with a 0. Similarly for PT002 and PT003. ID Sex Age Q1_1 Q1_2 Q1_3 Q1_4 PT PT PT ~ 8 ~

9 6. Cases and controls, or different groups of subjects: If you have, for example, cases and controls, put them all in the same spreadsheet. Use a column to record which are cases and which are controls. Do not put them in different places (e.g. different spreadsheets). Similarly if you have different groups of subjects, the same thing applies. Use one column to identify the group that a subject belongs to. ID Sex Age Group BMI Patient_type Walking_aid wheel rollator Walking stick Walking stick 7. ID numbers: Each case in the spreadsheet should have a unique identifier, so that errors in the spreadsheet can be traced and corrected. 8. Pair numbers (matched data only): If your study has matched pairs of cases and controls, each pair should have an identifying number. This is different from the unique identifier for each case. It is also different from the group identifier. It tells the statistics package which case and control make up a matched pair. In other words, you need two ID numbers, an ID number for each participant and an ID number for the pair that the participant belongs to. ~ 9 ~

10 9. Do not place text in columns that contain numbers: If a variable contains numbers, do not put extra text into it. For instance, do not write 'm' or metres to indicate metres. If you need to record the unit of measurement try to incorporate it into the variable name. Bad Example: Distance 22m metres m Good Example: Distance_m Units: Keep the same units throughout. For example, if you are recording minutes, don't put "1 hour 10 minutes" - convert this to minutes (70). The relevant unit for each variable should be recorded in the codebook/data dictionary (see section 4). Bad Example: Time 10 mins 1 hour 2 hours 5mins Good Example: Time_mins ~ 10 ~

11 11. Missing data: If information is missing just leave the cell blank. Do not put in things like 'NA' or 'Unknown' or?. In particular, do not use a numerical value such as 999 as this can be easily mistaken as non-missing. If you need to record why values in a particular variable are missing, put this information in a separate column. In the example below, a variable Missing_Age was used to record why Age was missing. It can be helpful to develop a numeric coding scheme describing reasons for missing values (see point 13 below). ID Sex Age Missing_Age PT PT002 2 Age illegible on patient form PT PT004 2 Age was not recorded 12. Writing numbers: Do not put commas into large numbers; write 2500, and not 2,500. And make sure you don't use the letters O and l when you should have 0 and 1. If you are recording the presence or absence of something, use 0 (zero) for absent and 1 for present. 13. Text variables: Text allows plenty of scope for misspelling and inconsistency. Don't use free text (e.g.: other: please specify) if you can avoid it. Develop a numerical coding scheme and classify the answers before entering the data. Where use of text is unavoidable, beware of capital letters: Depression and depression will appear as different things in your data. ~ 11 ~

12 14. Blank Lines/Empty Cells: A spreadsheet should never contain blank lines or empty cells unless the data is missing. Example 1 Bad Example: Cells were left blank where a date was supposed to be repeated several times - this should not be done under any circumstance. Good Example: No cells left blank, all completely filled in. ID Date Weight /05/ /05/ /05/ /05/ ID Date Weight /05/ /05/ /05/ /05/ /05/ /05/ /05/ /05/ /05/ /05/ ~ 12 ~

13 Example 2 Bad example: Day 1 Day 7 Beaumont Patient Pain Walking Distance Pain Walking Distance Connolly Empty cells: Appears as if the first recording of pain and walking distance was measured on day 1 and following this pain and walking distance was measured on day 7 again. Blank lines: between sites, Beaumont and Connolly (also data for sites should not be recorded in this manner) ID numbers: Not unique Good example: Patient Site Pain_Day1 WalkDist_Day1 Pain_Day7 WalkDist_Day7 1 Beaumont Beaumont Beaumont Connolly Connolly Connolly ~ 13 ~

14 15. No colour or highlighting: Do not use colour or alternatively highlight cells to indicate unusual values, different groups of people etc. For possible unusual values it would be better to have another column to indicate this. Similarly, for various groups of people, include a separate column to distinguish the various groups. 16. No calculations or graphs: Your primary data file should only contain data. You should not start calculating, for example, means of various variables, or insert graphs to look at trends. If you want to do this, make a copy of your final data file and do the calculations there. Finally, if in doubt about how to record your data in a spreadsheet, ask the DSC. You can always put in a couple of cases and attend one of the scheduled clinics to make sure it makes sense. Details of the scheduled clinics can be found on the DSC webpage: 5. Codebook / Data dictionary A codebook or data dictionary is a separate file (e.g. word file, excel spreadsheet) that explains what all the variables are. It contains information that helps users decipher data file content and structure and is extremely useful if data is being shared. A codebook should contain: The variable name as in the data file A small description or explanation of what the variable is if not self-explanatory Coding schemes Units Anything else you think might be useful (Filename(s), File location(s), brief description of study design and research questions) ~ 14 ~

15 Example Variable Name Descriptions Variable Type Code Units/Format Notes Age Age Continuous years Age should be >65 years only. Sex Sex Categorical 1=Female; 2=Male GPvisit Hospital Group Time Number of GP visits in the last 6 months. Number of hospital visits in last 12 months Whether the patient was in the intervention or control group Time to complete observations Categorical Categorical Categorical Continuous 1 = intervention; 2 = control minutes 6. Backups Always make back-ups of your data in multiple locations. Keep all versions of your data, in case you accidently delete something or the file is corrupt, or even worse, your computer breaks or is stolen. Before updating your data file each time you enter new data, make a copy and enter the data in the new version. ~ 15 ~

16 7. Checking your data data cleaning Once your data is entered (or part of your data entered) you should conduct a few checks to ensure that the data is correct. This is known as data cleaning. Data cleaning is the process of detecting and correcting (or removing) corrupt or inaccurate records (i.e. checking data entry). Identify incomplete, incorrect, inaccurate, irrelevant data and then replace, modify, or delete this data. As mentioned above, always store a copy of the original data. If you make any changes to the data file save it as a new file. Steps involved: 1. Define and determine error types and instances 2. Correct the errors 3. Document error instances and error types in a separate file. 4. If data entry is not complete, modify data entry procedures to reduce future errors. Examples of errors to check for: Extreme values, examples include an age of 130 the number of GP visits in a week = 56 the number of children =20 Inconsistencies, for example sex = male and hysterectomy = yes. Missing values - check that missing values are in fact missing and that they have not been omitted accidently. Duplicate Records - Check that records are not entered twice. How to check for errors It is relatively easy to visually check if you have a small data file. If you have a large data file there are two simple approaches to help identify invalid records in data set: frequency tables - to list all the unique values of a variable ~ 16 ~

17 summary / descriptive statistics listing means, ranges, minimum and maximum values Additionally, if you are familiar with excel you can use basic functions (e.g. conditional formatting, IF statements) to explore data and detect errors. Conditional formatting and IF functions can be used to highlight cells that do not conform to certain criteria. Example: This experiment was conducted on 10 patients aged years of age. Two treatments were investigated. ID Age Sex Treatment 1 46 male female female female male male female female mael male 2 Potential errors / inaccurate data: An ID of 44 when there are only 10 patients. Study is supposed to be conducted on patients aged years old but patient 5 is apparently 25. The variable Sex = mael for patient 9. There are only supposed to be two treatments but patient 9 received treatment 3. Where possible, check the errors and correct them. Thoroughly document any material changes made to the raw data file (e.g. removal of outliers) in a separate file to be shared with future collaborators. Always retain a copy of the original raw data for reference. ~ 17 ~

18 8. Data protection The Data Protection Act 1998 (DPA) protects individuals personal data and sets the standards for the processing of this information. It does this by requiring those processing personal data to comply with the eight data protection principles. 1. Obtain and process information fairly 2. Keep it only for one or more specified, explicit and lawful purpose 3. Process it only in ways compatible with these purposes 4. Keep it safe and secure 5. Keep it accurate, complete and up-to-date 6. Ensure that it is adequate, relevant and not excessive 7. Retain it for no longer than is necessary for the purpose or purposes 8. Give a copy of his/her personal data to that individual, on request All staff and students who are using personal data should comply with the provisions of the Act, including the eight principles above. Please see the following for more information on data protection at RCSI: Anonymisation and Pseudonymisation Before data obtained from research with individuals can be shared with other researchers, the DSC, or archived, you may need to anonymise the data so that individuals cannot be identified. Please note that there should be no reason to supply the DSC with data which includes information such as patient names, addresses etc. Any such data sent to the DSC will be returned to the researcher. Anonymisation This is the removal of information that could lead to an individual being identified. Data can be considered anonymised when it does not allow identification of the individuals to whom it relates, and it is not possible that any individual could be identified from the data, either by any further processing of that data or by processing it together with other information which is available or likely to be available. Identifiers can be direct, like the individual s name, or indirect, like their ~ 18 ~

19 date of birth, phone number and address. Hence, removing direct identifiers does not render data anonymous. Generally, information is fully anonymised if there are at least 3-5 individuals to whom the information could refer. Pseudonymisation Pseudonymisation should be distinguished from anonymisation. Pseudonymisation is a procedure by which the most identifying fields within a data record are replaced by one or more artificial identifiers, or pseudonyms. However, while pseudonymisation has many uses, unlike anonymisation, it only provides a limited protection for the identity of individuals and in many cases allows for identification using indirect means. 9. Useful resources RCSI Data Protection and Data Storage: EU General Data Protection Regulation: Data Protection - Anonymisation and Pseudonymisation: Data organization in spreadsheets: How to share data for collaboration: ~ 19 ~

20 Appendix A Saving your Excel spreadsheet as a CSV file 1. In Excel, click on File and choose Save As. ~ 20 ~

21 2. Under Save as type, choose CSV (Comma delimited). Click Save. 3. You may see a message "The selected file type does not support workbooks that contain multiple sheets." This message is to inform you that any additional sheets (e.g. Sheet2) that you may have information on will not be kept in this format (i.e. all information on the other sheets will be lost). Ensure that you don t have information on other sheets that you need and click OK to continue. ~ 21 ~

22 4. You may also see a message that your file "may contain features that are not compatible with CSV (Comma delimited). Do you want to keep the workbook in this format?" This message is to inform you that any formatting you may have (e.g. colours, lines, bold text), and / or any formulas will not be preserved in this format. Click Yes to continue. ~ 22 ~