What Is Scale in SPSS and Quantitative Research

What Is Scale

In quantitative research, a scale is usually understood as a measurement system used to represent a characteristic of an object through a variable or a set of levels. When you open Variable View in SPSS, you will see the Measure column with three main choices: Nominal, Ordinal and Scale. Many students call all three choices measurement scales in SPSS, although, in measurement terms, Scale in SPSS usually refers to continuous or approximately continuous quantitative variables.

You can think of a scale as the rule that turns a concept into data for analysis. The concept of gender is coded into groups, satisfaction is recorded in an order from very dissatisfied to very satisfied, and age is recorded as a number with a clear measurement distance. The measurement level determines which comparisons, calculations and tests you can run on that variable.

In a Likert questionnaire, each item usually has levels from 1 to 5 or from 1 to 7. The Likert scale was introduced in Likert's work (Likert, 1932). When entering data, you may set each item to Ordinal in theoretical terms. However, in many business studies, Likert items are combined into a mean score for a construct and then entered into regression or PLS-SEM as a Scale variable. Your decision needs to be consistent with the method and with your supervisor's requirements.

If you are reviewing other topics on data processing, you can open the Statistics category to compare this article with posts on hypothesis tests and regression.

Why Scale Matters in Quantitative Research

Choosing the correct measurement level helps you determine how the data can be processed. Nominal variables are used to form groups and count frequencies. Ordinal variables allow you to rank levels, but the distance between level 1 and level 2 may not be the same as the distance between level 3 and level 4. With Scale variables, means, standard deviations and many statistical models are usually more appropriate.

For example, the variable department may have the values 1 = Economics, 2 = Marketing, and 3 = Finance. The numbers 1, 2 and 3 are only group codes. They do not mean that Finance is higher than Marketing. If you set this variable to Scale, SPSS can still calculate the Mean, but that result has no research meaning.

By contrast, monthly income can be a Scale variable if you enter the actual amount of money. An income-level variable divided into below 5 million, from 5 million to below 10 million, and 10 million or more is usually Ordinal because the groups have an order, but the distances between the groups are not equal in the same measurement unit.

The measurement level also affects your choice of charts and tests. Categorical variables are commonly used with Frequency, Crosstabs or Chi-square. Quantitative variables can be used with Descriptives, Correlations, Linear Regression and mean comparison tests when the relevant conditions are met. Before running an analysis, check both the data type in the Type column and the measurement level in the Measure column.

Do not confuse Scale in the Measure column with the scale of a construct. A satisfaction scale may contain SAT1, SAT2, SAT3 and SAT4. Each item is entered as a column, and you then assess Cronbach's Alpha, EFA or the measurement model. The Measure column only describes how SPSS treats the variable. It does not prove that your scale is reliable.

What Value of Scale Is Acceptable

There is no single value that makes Scale in the Measure column acceptable. You need to consider the variable group, the measurement unit and the analysis for which the variable will be used. The thresholds below apply to common situations in a quantitative thesis. They do not replace reading the full model.

The table contains two different groups of information. The first four lines concern reliability and EFA in SPSS. The final two lines are used to assess the measurement model in PLS-SEM. You should not use AVE or Composite Reliability to draw a direct conclusion about a Cronbach's Alpha table in SPSS.

When reading a threshold, consider the context as well. A new scale may be accepted with an Alpha of 0.6 in exploratory research according to (Hair et al., 2010), but you need to explain the reason and report the Item-Total Statistics results. Automatically deleting an item simply because Alpha is below the preferred level can remove important content from the construct.

How to Read Scale in the Output

Suppose you have four items, SAT1 through SAT4, measuring satisfaction on a Likert scale from 1 to 5. You run Analyze > Scale > Reliability Analysis, move the four variables into the Items box, select Alpha as the Model, then click Statistics and select Scale if item deleted. The table below is illustrative output, not the result of a real study.

In the Reliability Statistics table, the illustrative Alpha is 0.872. This exceeds the 0.7 threshold commonly used as a reference for internal reliability (Nunnally, 1978). However, SAT3 has a Corrected Item-Total Correlation of 0.276, which is below 0.3 according to (Nunnally and Bernstein, 1994). This is a signal to investigate, not an automatic instruction to delete the item.

The Cronbach's Alpha if Item Deleted column shows how Alpha would change if the item were removed. In this example, deleting SAT3 increases Alpha from 0.872 to 0.901. You still need to read the wording of SAT3, check whether it was reverse-coded, consider whether respondents understood the item, and compare it with the original scale before deciding.

If you run EFA, the common tables are KMO and Bartlett's Test, Total Variance Explained and Rotated Component Matrix. KMO indicates whether the data are suitable for examining a factor structure. In the Rotated Component Matrix, inspect the factor loading of each item and check whether an item loads strongly on more than one factor.

If you analyse the data in SmartPLS, the table names will be different. You may see Outer Loadings, Construct Reliability and Validity, Discriminant Validity and Path Coefficients. In that situation, Scale in SPSS is not the main criterion for evaluating a PLS-SEM model. You assess outer loading, Composite Reliability, AVE and HTMT using the relevant criteria.

What to Do When Scale Results Are Not Acceptable

Start by checking the original data. Open Variable View and review Label, Values, Missing and Measure. Then go to Data View and look for blank cells, values outside the Likert range, or codes such as 99 or 999 that were entered as valid responses. A low Alpha may sometimes result from a data-entry error rather than a poor scale.

Next, check reverse-coded items. If the questionnaire contains negatively worded statements, you need to create the reverse-coded variable before running the reliability analysis. For a Likert scale from 1 to 5, the usual coding is new value = 6 minus old value. Record this operation in the data file and give the new variable a clear name so that you do not overwrite the original data.

Then review Item-Total Statistics. You may remove an item when its Corrected Item-Total Correlation is low, its content does not fit the construct, or it creates cross-loading in EFA. The decision should be based on both theory and the output. Running the analysis repeatedly to find the highest Alpha by deleting items in batches can damage content validity and leave you with a result that is difficult to defend.

If several items cause problems, check the questionnaire translation, the response instructions and the quality of the sample. You should also review outliers in SPSS when the data contain cases with unusually different responses. An outlier is not automatically an incorrect response, so any deletion needs a clear criterion and a record in your analysis log.

If the scale still does not meet the criteria after these checks, you have three reasonable options. You can retain the scale and explain that the study is exploratory if the indicators remain acceptable. You can remove an item with a clear justification and rerun the analysis in the correct order. If the questionnaire or data contain a serious error, discuss additional data collection with your supervisor instead of trying to repair the result through software.

Distinguishing Scale from Other Measurement Levels

The four levels commonly discussed are Nominal, Ordinal, Interval and Ratio. In SPSS, the Measure column usually displays Nominal, Ordinal and Scale. Interval and Ratio are generally grouped under Scale because both are quantitative variables, although their measurement meanings differ.

Nominal is used for groups with no natural order. Examples include gender, field of study, province and customer type. Codes 1, 2 and 3 are labels only. You can count the number of cases in each group, calculate proportions or create dummy variables, but you should not calculate the mean of the group codes.

Ordinal has an order, but the distances between levels may not be equal. Satisfaction rankings, education levels and income groups are familiar examples. You know that level 4 is higher than level 3, but you cannot assume that the difference between 4 and 3 equals the difference between 2 and 1.

Interval has equal distances between levels, but zero does not necessarily represent the complete absence of the measured attribute. Celsius temperature is a commonly discussed example. Ratio has equal distances and an absolute zero, such as the number of products, years of work experience or revenue.

With a Likert questionnaire, the discussion often concerns whether each item should be treated as Ordinal or Scale. Follow the research design and your discipline's guidance. If you use the mean of several items to represent a construct, describe clearly how the score was calculated and report the reliability assessment completed beforehand.

Scale is also easily confused with std, the abbreviation for standard deviation. Scale is the variable's measurement level in SPSS, while standard deviation describes the dispersion of the data. You can also read what std means to avoid confusing these two terms in the Descriptive Statistics table.

Common Errors

Setting every variable to Scale. This is the quickest error to make during data entry. Gender codes, field-of-study codes and area codes remain categorical variables even when they are entered as numbers. Set Measure according to the meaning of the variable, not according to whether the data cell contains text or numbers.

Setting every Likert item to Nominal. This ignores the order of the response levels and may create problems when you choose an analysis. Decide whether you are analysing each item separately or using a composite score for the scale.

Deleting an item as soon as Alpha increases. A higher Alpha alone does not prove that the scale is good. Read the Item-Total Statistics, the item wording, the EFA results and the theoretical model.

Using one threshold for every analysis. Factor loading in EFA, outer loading in PLS-SEM, AVE and CFI belong to different contexts. In particular, CFI, TLI, RMSEA and SRMR according to (Hu and Bentler, 1999) belong to CB-SEM and should not be used to conclude a SmartPLS analysis.

Failing to save each data version. When you delete SAT3 and rerun the analysis, save the syntax or keep a log containing the variable name, the reason for deletion, and Alpha before and after the change. This allows you to answer why an item is absent from the final model.

Confusing a moderator with a mediator. These are concepts in a relationship model, not measurement levels. If your study includes a moderating variable, you can read what a moderator is and distinguish it from a mediator before running the model.

Frequently asked questions

What is Scale in SPSS?

Scale in SPSS is an option in the Measure column, usually used for quantitative variables that can be calculated, such as age, income, mean scores or total scale scores. It differs from the scale of a latent construct in a research model.

How many basic measurement levels are there in SPSS?

The SPSS Measure column has three main choices: Nominal, Ordinal and Scale. In measurement theory, many sources also distinguish Interval and Ratio. SPSS usually groups Interval and Ratio under Scale.

Should a Likert variable be Nominal, Ordinal or Scale?

Each Likert response level has an order, so it can be treated as Ordinal. In many quantitative studies, the mean score of several Likert items is treated as Scale for regression or later analyses. State this convention clearly in your research methods section.

What Cronbach's Alpha is acceptable?

An Alpha of 0.7 or above is commonly used as a reference for scale reliability according to (Nunnally, 1978). For a new scale or exploratory research, an Alpha of 0.6 may be considered according to (Hair et al., 2010). You also need to review the Corrected Item-Total Correlation and the item content.

Does an unacceptable Scale result mean that I must delete an item?

Not necessarily. Check reverse coding, missing data, incorrect values and item content first. If you remove an item, the decision should have support from both the indicator and the theory. Then rerun the analysis and save the processing history.

Does the Measure column determine the SPSS result?

Measure helps SPSS and the analyst identify the measurement level, but it does not automatically prove that the data meet the requirements of a test. You still need to check assumptions, scale quality, coding and the objective of the model.

How to write this in your thesis

Open your data file now and create a list for each variable that includes its name, meaning, coding and Measure. Then review the item groups before running Cronbach's Alpha. If you need to run the analysis on your own .sav or .csv file, DoThesis M4 analysis handles this step.