Introduction

Analysis begins long before the first table is produced. The fictional Cedar Valley Social Connection Study, which this course follows, has finished collecting its regional survey of adults aged 65 and older: 1,600 people completed it, and 392 of them (24.5 percent) scored 6 or higher on the three-item UCLA Loneliness Scale. Before the team can report those numbers with confidence, someone has to make sure that the file holding them is complete, correctly laid out and fully documented. This section covers the three habits that prevent most of them: entering data in a consistent layout, writing a data dictionary, and handling missing values deliberately. It then applies these ideas to a real Canadian dataset that Section 2 analyzes.

1.1 Data Entry and the Shape of a Dataset

One row, one column, one value

Almost every statistical program expects data in a rectangular layout. Each row holds one unit of observation, which in a survey is usually one participant. Each column holds one variable, which is a characteristic that can take different values, such as age or loneliness score. Each cell holds one value. A unique participant identifier in the first column links each row to the consent record and, later, to linked administrative data (Lesson 9), without storing the person's name in the analysis file. Broman and Woo (2018) give practical rules for spreadsheets that follow from this layout: use one header row with short names, put nothing but data in the data cells, never use colour or bold type to carry information, keep codes consistent (always "Woman", never a mix of "W", "woman" and "F"), write dates in the year-month-day form (2026-03-14), and avoid merged cells and blank rows.

The table above shows four illustrative rows of the Cedar Valley raw file. The numbers in the gender, community and UCLA columns are codes, and a reader cannot interpret them without the data dictionary in Section 1.2. The value -88 for participant CV-04580 is a missing code meaning that the person chose not to answer the third loneliness item.

Entering paper forms

Typing paper forms into a file introduces errors such as transposed digits, skipped lines and values entered in the wrong column. The standard protection is double data entry: two people enter the same forms independently, a program compares the two files cell by cell, and every disagreement is checked against the paper original. A second protection is a set of validation rules (also called range checks) that refuse impossible values at the moment of entry, such as an age of 6 in a study of people aged 65 and older or a UCLA item score of 4 on a scale that runs from 1 to 3. Electronic capture in REDCap or Qualtrics applies these rules as participants answer (Lesson 8), although exported files still need checking because a rule only catches the errors someone anticipated.

Protecting the raw file

The raw export is the study's primary record. The team saves it once, marks it read-only, and never edits it. Every change, from fixing a typing error to creating a new variable, is made by code in a script that reads the raw file and writes a separate analysis file. Anyone can then rerun the script to reproduce the analysis file, and a cleaning mistake can be found and reversed, in keeping with the file organization habits of Lesson 6.

1.2 Codebooks and Data Dictionaries

A data dictionary is a table that describes every variable in a dataset. For each variable it records the variable name used in the file, a variable label that gives the full question or a plain description, the type of data, the permitted values and what each code means (the value labels), the missing codes, the units, and the source or derivation. The word codebook is often used for the same document, and some codebooks also add the frequency of each value. Section 3 uses "codebook" in a different sense, for the list of qualitative codes applied to interview text, so this lesson keeps the two terms apart by calling the quantitative document a data dictionary. Lesson 3 began a variable list from the Cedar Valley DAG, and Lesson 8 added the survey items; the data dictionary is where that list becomes complete.

Two entries deserve attention. The derived variables ucla_total and lonely do not appear on the survey; they are created in the cleaning script, and the dictionary states the rule that creates them so that anyone can check it. The three UCLA items are each scored from 1 to 3 (Hughes et al., 2004), so the total runs from 3 to 9, and the Cedar Valley team decided in its protocol that a score of 6 or higher would count as lonely.

HSCI 410 Lesson 1 Section 2 builds a full study data dictionary on these principles, and HSCI 410 Lesson 2 Section 1 uses the data dictionary to set valid ranges for data cleaning.

1.3 Missing Values and Missing Codes

A value can be missing for different reasons, and the reasons matter for analysis. A question may be not applicable because branching logic skipped it (a question about a spouse's health is skipped for someone who is not married). A participant may decline to answer or answer that they do not know. A participant may stop before reaching the question, or may never be shown it. Many datasets record these reasons with numeric missing codes such as -77, -88 and -99, which keep the reasons distinct in the raw file.

Numeric missing codes become dangerous when they are treated as real numbers. Suppose four Cedar Valley participants report ages of 72, 68, 81 and 75, and a fifth has the code -99. The mean of the four real ages is 74.0 years. If the -99 is left in, the computed mean of the five values is 39.4 years, which looks like a plausible adult age and could pass unnoticed. Before any analysis, the cleaning script must convert each missing code to the value that the software recognizes as missing. Many programs display that value as NA (not available). Summaries such as the mean are then calculated from the recorded values only, and the number missing is reported beside them, as Section 2 shows.

1.4 A Real Dataset: the Canadian Social Connection Survey

The Cedar Valley data are fictional, so the series practises on a real dataset. The Canadian Social Connection Survey (CSCS) is a national online survey of social connection, loneliness and health among people living in Canada, and a de-identified public version is available on GitHub. The public file combines several survey years, and this lesson uses the 2021 wave. The worked examples below show what the file contains and how its data dictionary and missing values look in practice.

Reading the names and the dictionary

With 3,247 columns, nobody memorizes the variables. The CSCS names begin with a prefix for each part of the survey, and the metadata record the question text for each variable in each wave.

Inspecting a variable and finding its missing codes

Introduction

Descriptive statistics summarize who is in a dataset and what their answers look like. They come before any comparison or model, for two reasons. First, they let the team check the data: an age of 340 or a loneliness score of 12 shows up immediately in a frequency table. Second, readers need to know who was studied before they can judge whether the findings apply to anyone else. In most health research articles this description appears as the first table, which is why it is called Table 1. This section works through frequency tables, numeric summaries and cross-tabulations, and then builds a Table 1 from the CSCS 2021 data for respondents aged 65 and older, the age group that the fictional Cedar Valley study recruits.

2.1 Matching the Summary to the Variable

Lesson 7 introduced levels of measurement. They decide which summary is appropriate. Categorical variables, whether nominal (categories with no order, such as community of residence) or ordinal (ordered categories, such as self-rated health from poor to excellent), are summarized with counts and percentages. Numeric variables measured on an interval or ratio scale, such as age or the number of emergency department visits, are summarized with a measure of centre (the mean or median) and a measure of spread (the standard deviation or the interquartile range).

HSCI 410 Lesson 2 Section 3 extends these summaries to measures of shape, including skew, and to transformations of skewed variables.

2.2 Frequency Tables

A frequency table lists each value of a variable with the number of participants who gave it. Adding percentages makes groups of different sizes comparable. The Cedar Valley team reports that 392 of 1,600 respondents, or 24.5 percent, scored 6 or higher on the UCLA scale. A percentage always has a denominator, and the analyst must say what it is. A valid percentage uses only the participants with a recorded answer as the denominator, while a percentage of all participants includes those with missing values. The difference is small when little is missing and large when much is missing.

The CSCS analysis starts by keeping respondents aged 65 and older, so that the practice data resemble the Cedar Valley population.

2.3 Describing Numeric Variables

The mean is the sum of the values divided by their number. The median is the middle value when the values are sorted, so half of the participants lie below it and half above. The standard deviation (SD) describes how far values typically lie from the mean: a small SD means that most values are close to the mean, and a large SD means that they are spread out. The interquartile range runs from the first quartile (the value below which a quarter of participants lie) to the third quartile (the value below which three quarters lie), and it contains the middle half of the data.

The mean and the median agree when values are spread symmetrically, and they part ways when a few values are extreme. Suppose nine Cedar Valley participants had 0, 0, 0, 0, 1, 1, 2, 3 and 11 emergency department visits in a year. The total is 18, so the mean is 2.0 visits, while the median is 1 visit. One person with 11 visits pulls the mean upward, and the median better describes the typical participant. Counts of health service use are often skewed in this way, which is why the table in Section 2.1 recommends the median for them.

2.4 Cross-Tabulations

A cross-tabulation (or two-way table) counts participants in every combination of two categorical variables. It is the first step toward asking whether two characteristics go together. Before tabulating, any missing code stored as a category must be converted to NA, or it will appear as a real group in every table.

HSCI 410 Lesson 2 Section 3 returns to frequency tables and cross-tabulations and adds the odds ratio to the cross-tabulation.

2.5 Building a Table 1

A Table 1 describes the analytic sample, the participants included in the analysis, usually overall and separately for the groups that the study compares. The STROBE reporting guideline for observational studies asks authors to give the characteristics of study participants and the number with missing data for each variable (von Elm et al., 2007), and Lesson 12 returns to STROBE and to table design. Building the table involves five decisions, which the cards below summarize.

Table 1. Characteristics of CSCS 2021 respondents aged 65 and older, overall and by score on the three-item UCLA Loneliness Scale. Percentages are column percentages among respondents with a recorded value. Of 486 respondents aged 65 and older, 53 without a UCLA score were excluded. Loneliness scores range from 3 to 9; a score of 6 or higher was classed as lonely.

Reading a Table 1

The table describes the analytic sample. Mean age is the same in both groups. Respondents classed as lonely were more often single and not dating (46.3 percent compared with 32.2 percent) and more often rated their physical health as poor (10.3 percent compared with 1.1 percent) or fair (27.0 percent compared with 17.2 percent). These are descriptions of differences in this sample. The table contains no statistical tests, and on its own it cannot show whether the differences would hold in other samples or whether loneliness and health influence one another. Statistical tests are taught in HSCI 341 Lesson 6 Section 4 and HSCI 410 Lesson 3 Section 1, regression models in HSCI 410 Lesson 3, and the causal questions require the reasoning of HSCI 341. The table also shows the analyst's choices openly: the row order runs from the most to the least favourable health rating, the overall column comes first, and the footnote gives the denominator and the exclusions.

Introduction

Part 2 of this lesson turns from numbers to words. The qualitative strand of the fictional Cedar Valley Social Connection Study includes 24 semi-structured interviews with older adults living alone and four focus groups: two with older adults, one with family caregivers, and one with clinic staff and community connectors. Lesson 10 prepared a clean verbatim transcript of one interview. A single interview of 54 minutes can produce fifteen or more pages of text, and 24 of them produce several hundred. Coding is the first systematic step in making that volume of text manageable without losing what participants said. This section builds a starter codebook for the Cedar Valley interviews and uses it to code the excerpt from Lesson 10. It gives a working method for a first coding pass, and HSCI 841 Lessons 5 to 12 develop qualitative analysis in depth.

3.1 What a Code Is

A code is a short label, usually a word or phrase, that a researcher attaches to a passage of text to record what the passage is about or what it means for the research question (Saldaña, 2021). The passage is a coded segment; it may be a phrase, a sentence or a few sentences that express one idea. One segment can carry more than one code, and the same code is applied to every segment, in every transcript, that fits its definition. Coding lets the researcher gather every segment with the code "transportation" from all 24 interviews and read them side by side.

Saldaña (2021) distinguishes first-cycle coding, the initial labelling of segments, from second-cycle coding, in which codes are grouped, compared and organized into larger categories and themes. A theme is a pattern of meaning that runs across many segments and participants and says something about the research question. This lesson stops at first-cycle coding. Section 4 shows how different analytic approaches carry codes forward. HSCI 841 Lesson 5 Section 1.4 teaches the move from codes to themes, and HSCI 841 Lesson 6 turns codes into conceptual models.

Coding goes together with writing memos. A memo is a short dated note in which the researcher records an idea, a question, a possible connection between codes or a reaction to the data. Analytic memos written during coding continue the reflexive memos of Lesson 10 and often hold the first drafts of later themes.

3.2 Deductive and Inductive Codes

Codes come from two directions. Deductive codes (also called a priori codes) are written before coding begins. They come from the research question, the topics of the interview guide, the causal web drawn in Lesson 3, or an existing theory. For example, Weiss (1973) distinguished emotional loneliness, the absence of a close attachment figure, from social loneliness, the absence of a wider network of friends and acquaintances, and a team could write a code for each. Inductive codes are created during coding, when a passage expresses something important that no existing code captures. One useful kind is the in vivo code, which uses the participant's own words as the label, such as "a room full of strangers". Most applied health studies combine the two, starting with a short deductive list and adding inductive codes as the data require. Fereday and Muir-Cochrane (2006) describe this hybrid approach and the record-keeping it needs.

3.3 Building a Starter Codebook

A qualitative codebook is the list of codes with the rules for applying them. MacQueen and colleagues (1998), writing about team-based analysis, recommended that each entry give a short code name, a brief definition, a full definition, guidance on when to use the code, guidance on when not to use it, and an example. The "when not to use" rule is the part beginners most often leave out, and it is the part that keeps two similar codes apart. A starter codebook holds the deductive codes written before coding begins; it is version 1.0, and it will change.

The Cedar Valley team wrote seven deductive codes from its SPIDER question (how older adults living alone experience social connection after a move to a smaller town), the interview guide and the causal web.

The table shows four of the six columns that MacQueen and colleagues recommended. The full codebook also has a brief definition for quick reference and an example segment for each code, which the team adds as soon as it finds a clear example in the data. HSCI 841 Lesson 5 Section 2.4 extends the six columns to seven elements by splitting the example into positive and negative examples.

3.4 Coding a Transcript Step by Step

A first coding pass

The table shows one coder's first pass. Deductive codes are in capitals; proposed inductive codes are in italics, with in vivo codes in quotation marks.

The pass proposes four inductive codes: Chance encounters, Concealing loneliness, Personal invitation and Small-town visibility. Each now needs a definition and "use when" and "do not use when" rules before it enters version 1.1 of the codebook. For example, Personal invitation could be defined as "a direct, individual request from a person to join an activity, described as affecting whether the participant attended", to be used when the participant links attendance to being asked, and not used for general publicity such as notices. One interview cannot show whether these codes recur; the memos flag them so the coders can watch for them in the remaining 23 interviews.

3.5 Coding as a Team, and Tools for Coding

When more than one person codes, the team agrees on the codebook first. Two coders then code the same two or three transcripts independently, compare their coding segment by segment, and discuss each disagreement. Disagreements usually reveal an unclear definition, which the team rewrites. Some teams then calculate a measure of intercoder agreement such as Cohen's kappa, while others, particularly those using reflexive approaches, treat discussion itself as the check. HSCI 841 Lesson 5 covers agreement measures and when they suit an approach.

Coding can be done with highlighters on paper, comments in a word processor, or a spreadsheet with one row per segment, as in the table above. Qualitative data analysis software stores the transcripts, codes and memos together and retrieves every segment with a given code in one step. Taguette is a free, open-source option that HSCI 841 uses; NVivo, ATLAS.ti, MAXQDA and Dedoose are commercial programs. Whatever the tool, transcripts must be de-identified before coding, and a cloud-based program may only be used if the consent form and the data management plan permit it (Lessons 5 and 10).

Introduction

Coding is common to most qualitative work, but what happens to the codes afterwards depends on the analytic approach. An approach is a recognized way of moving from data to findings, with its own aims, procedures and standards of quality. Thematic analysis looks for patterns across participants; interpretative phenomenological analysis stays close to each person's experience; narrative analysis keeps stories whole. Ideally the approach is chosen when the study is designed and written into the protocol (Lesson 6), because it affects how many participants are needed, how interviews are run and how transcripts are prepared. This section introduces seven approaches used in health research, maps them to the data collection methods of Lesson 10, and shows how the fictional Cedar Valley team chose among them. It is an orientation. HSCI 841 Lesson 1 Section 4 relates the quality criteria in Section 4.3 to systematic, transparent and replicable analysis, and Section 4.4 lists the HSCI 841 lesson for each approach.

4.1 Seven Analytic Approaches

Each card below gives the aim of the approach, the data it usually works with, its main steps and a foundational source.

The approaches share some steps. Each begins with close reading, and most involve coding. They differ in what they produce: a set of themes, a set of categories with frequencies, a comparison matrix, a rapid summary for decision-makers, an explanatory theory, a detailed account of a few people's experience, or an analysis of stories. They also differ in how they treat a codebook. Framework analysis, rapid analysis, directed content analysis and codebook forms of thematic analysis use a structured codebook like the one built in Section 3. Reflexive thematic analysis, grounded theory and interpretative phenomenological analysis develop codes more freely and treat a fixed codebook with caution.

What each approach would do with Ruth's interview

The coded excerpt from Section 3 shows how the same first-cycle codes lead in different directions. The table describes what an analyst using each approach would do next with interview CV-INT-14.

4.2 Matching Approaches to Data Collection Methods

Lesson 10 introduced in-depth interviews, key informant interviews and focus groups. Two other common sources of qualitative data are open-ended survey questions and documents such as policies, meeting minutes or program records. The table maps the seven approaches to these sources. "Strong fit" means that the approach was designed for that kind of data or is routinely used with it; "possible" means that it is used with that data with some adaptation; a blank cell means that the combination is uncommon.

Two patterns stand out. Interpretative phenomenological analysis and narrative analysis need long, personal accounts from individuals, so they depend on one-to-one interviews (or, for narrative analysis, diaries and written life stories), and focus groups suit them poorly because the group conversation interrupts each person's account. Qualitative content analysis and thematic analysis can handle short answers from many people, which makes them the usual choices for open-ended survey questions. The table is a guide drawn from common practice; methodologists debate some cells, and HSCI 841 examines those debates.

4.3 Choosing an Approach

Four questions narrow the choice. What does the research question ask for: patterns, counts, comparisons, an explanation of a process, the depth of individual experience, or stories? What data will there be, and how much? When are findings needed, and by whom? Who will do the analysis, and how many people? The diagram pairs typical answers with the approaches that suit them.

The Cedar Valley decision

The Cedar Valley team considered several options for its 24 interviews and four focus groups. The tabs summarize its reasoning, which the team recorded in a memo and then in the protocol.

4.4 Where HSCI 841 Picks Up

HSCI 841, the series' graduate course in qualitative research methods and analysis, assumes the skills of this lesson: preparing transcripts, building a codebook and completing a first coding pass. The table shows where each approach is developed.

Bringing It All Together

This lesson took the first analytic steps with both kinds of data that the fictional Cedar Valley Social Connection Study collects. Part 1 prepared survey data for analysis: a rectangular layout, a read-only raw file, a data dictionary and deliberate handling of missing codes. It then used the 2021 wave of the Canadian Social Connection Survey to work through frequency tables, descriptive statistics, cross-tabulations and a Table 1 for respondents aged 65 and older.

Part 2 prepared interview data for analysis. It defined codes, segments and memos, distinguished deductive from inductive codes, built a starter codebook with rules for when to use and not use each code, and coded an excerpt from the interview with Ruth. It then described seven analytic approaches, matched them to the data collection methods they suit, and showed how the Cedar Valley team chose framework analysis for its interviews and rapid analysis for an interim brief.

The two parts follow the same logic. In each, written rules (a data dictionary or a codebook) make the work repeatable, missing or unclear material is handled openly, and the first summary (a Table 1 or a coding table) prepares the ground for analyses taught in HSCI 410 and HSCI 841.

The final reflection asks you to compare the quantitative and qualitative steps of this lesson and to plan the first analytic steps for a new dataset and two new transcripts.

Final Knowledge Assessment

Search the terms, tools and people introduced in this lesson.