1
Professonalization of Exercise Physiologistsonline
Vol 6 No 4 April 2003
SPSSGuide for Exercise Physiology
Larry Birnbaum
Department of Exercise Physiology
The College of St. Scholastica
Duluth, MN55811
Introduction
This tutorial is intended to help EXP students engaged in research use SPSS to statistically analyze their data. It does not cover all SPSS capabilities, but rather focuses on what EXP students most likely need. Specific examples are provided to illustrate how to enter and analyze data and interpret the results.
Whenever the same subjects are used for both treatment and control groups, SPSS refers to this as a repeated measures design. Most EXP studies are likely to be repeated measures. Furthermore, since we typically measure several physiological parameters (i.e., dependent variables), we have a multivariate repeated measures design (ANOVA), and since we measure the physiological parameters at two or more times, it is a doubly multivariate repeated measures design.
Example 1
Five* subjects participate in a massage study. Each subject will rest for 15 minutes (control) or receive a 15 minute massage (treatment) followed by 15 minutes of exercise at 75% of HRmax. Whether the subject receives the massage or rest is randomly determined, but each subject is used for both groups (massage; no massage). For example, subject 1 receives massage then exercises for the first round of testing. The next time she comes in for testing, say 2 days later, she rests for 15 minutes, then exercises. Subject 2 might not get the massage for the first round depending on the coin toss, or whatever method is used to randomly decide who gets the treatment (massage) for the first round of testing. He would rest for 15 minutes then exercise for the first round. The next time he came to the lab for testing, he would receive the massage, then exercise. Several physiologic variables are measured during (e.g., VO2, VCO2, VE, Fb, HR) or near the end of each session (e.g., BP, Q). The question is, “ Does massage significantly alter any physiologic parameters at rest or during exercise?” The null hypothesis states that massage does not alter any physiologic parameters at rest or during exercise. If this was a dissertation, there would be a null hypothesis written for each dependent variable.
*This example is for illustration only. A minimum of 10-12 subjects should be used in an EXP study. Numerous references state that 20 subjects is the minimum to achieve sufficient power, validity, etc., but completing an EXP study in one semester generally necessitates fewer subjects.
Example 2
A tilt-table study is performed on five subjects. Several physiologic variables are measured in three positions, upright, supine and head-down. The subjects remain in each position for 10 minutes and measurements are made during and at the end of each 10 minute period. The null hypothesis states that none of the physiological variables will be significantly different in any of the three positions. The alternative hypothesis states that one or more physiologic variables will be significantly different in at least one position compared to the other two positions.
Entering Data
Example 1
For illustration purposes, only heart rate and systolic blood pressure will be used. To get SPSS started, click on Start > Programs > SPSS for Windows > SPSS 11.0 for Windows. The Data View will be displayed on the screen. You will be entering data in the cells, but you need to describe your data in the Variable View first. Click on the Variable View tab (lower left corner). The name cannot be more than 8 characters. For this example, we will use the following names for the heart rate and systolic blood pressure data:
hrcr = heart rate, control (no massage), at rest
hrmr = heart rate, massage, at rest
sbpcr = systolic blood pressure, control, at rest
sbpmr =systolic blood pressure, massage, at rest
hrce = heart rate, control, during exercise
hrme = heart rate, massage, during exercise
sbpce = systolic blood pressure, control, during exercise
sbpme = systolic blood pressure, massage, during exercise
The Type is Numeric, Width is typically set at 8, Decimals at 0 since HR and SBP are whole numbers, and the remaining columns are left at their default settings. The Width setting indicates the number of characters (including numbers, commas, letters, etc.). Use your Manuscript Guidelines handout to help you select the correct decimal setting for each physiologic (dependent) variable.
Once you have the data described in the Variable View, click on Data View and enter your data under the appropriate column headings. It is helpful to have another person read the data to you as you input it. Be careful to accurately input your data. Obviously, any input error will produce erroneous results. When all your data is entered, your are ready to analyze it. You should save it first by clicking on File > Save As, then specify where you want it saved (use a floppy disk) and give it a file name that you will remember (e.g., massagerawdata.sav). Note the extension is always “sav” for raw data.
Example 2
As in example 1, only two variables, stroke volume and systolic blood pressure, will be used to illustrate how to use SPSS. The names used for these two variables measured in each of the three positions will be as follows:
SVU = stroke volume, upright
SVS = stroke volume, supine
SVD = stroke volume, head down
SBPU = systolic blood pressure, upright
SBPS = systolic blood pressure, supine
SBPD = systolic blood pressure, head down
As in example one, the Type is Numeric and the Decimalsettingis set at 0 since SV and SBP are whole numbers. Once you have the data described in the Variable View, click on Data View and enter your data under the appropriate column headings. When all your data is entered, save it to a floppy disk, then analyze it.
Analyzing Data
Example 1
There is one within-subjects factor in this study, massage, which has two levels, massage vs no massage. We are measuring the effect of massage at rest and during exercise only to make the study more efficient. In essence, we are doing two separate studies, one at rest and one with exercise. Thus, while we input the data as four variables for each physiological measure (HR/no massage/rest, HR/massage/rest, HR/no massage/exercise, HR/massage/exercise), we will only be comparing two of the variables for each condition (rest and exercise).
Once your raw data is entered, click on Analyze > General Linear Model (GLM) > Repeated Measures. For the within-subject factor name, type “massage” in the window, then type in “2” for the number of levels. Click on the Add button. Now click on the Measure button. Type “hr” in the window for heart rate, then click on the Add button. Do the same for systolic blood pressure using “sbp” as the label. Repeat this for each physiological parameter measured. The screen should appear as follows:
Next, click on the Define button. Highlight the appropriate variables in the window on the left, then click on the arrow (►). Be sure each dependent variable is identified correctly (e.g., hrcr with (1, hr); sbpmr with (2, sbp)) in the Within-Subjects window. We will do two analyses, one that compares all the dependent variables for massage and no massage at rest, and one that compares all the dependent variables for massage and no massage during exercise. Once all the comparisons are defined (for rest), the screen should appear as below. Click on the OK button. It will take a few seconds to complete the analysis. Repeat the process to analyze the dependent variables during exercise. Save your output data to your floppy disk using a file name you will remember. Use the correct extension (spo). You may wish to print a hard copy as well. Print in the landscape mode.
Example 2
There is one within-subjects factor in this study, body position, which has three levels, upright, supine and head down. Once your raw data is entered, click on Analyze > General Linear Model (GLM) > Repeated Measures. It is a repeated measures design because we are using the same subjects for all three positions. For the within-subject factor name, type “position” in the window, then type in “3” for the number of levels. Click on the Add button. Now click on the Measure button. Type “sv” in the window for stroke volume, then click on the Add button. Do the same for systolic blood pressure using “sbp” as the label. Repeat this for each physiological parameter measured. The screen should appear as follows:
Now click on the Define button. Highlight the appropriate variables in the window on the left, then click on the arrow (►). Each variable must be “defined” correctly. We will do one analysis that compares all the dependent variables for each body position. Once all the comparisons are defined, the screen should appear as below. Click on the OK button. It will take a few seconds to complete the analysis. Save your output data to your floppy disk.
Interpreting the Results
Example 1
Check the first table of Within-Subjects Factors to verify that each dependent variable is identified correctly. In this example, note that in the first column no massage is identified as #1 andmassage as #2. This should correlate with your identification scheme in the Dependent Variable column (e.g., heart rate no massage at rest isHRCR = HR 1; heart rate massage at rest is HRMR = HR 2).
Now you have a lot of test results, but which ones do you use to interpret your data (i.e., are any of the measured physiologic parameters significantly affected by the treatment?). First, check to see if the sphericity assumption is met (Mauchly’s Test of Sphericity). Epsilon should be greater than 0.75 (1.000 indicates perfect sphericity). If epsilon is <0.75, the results of the Greenhouse-Geisser or Huynh-Feldt corrections (univariate tests) may be used. If the sphericity assumption is met, we are interested in the Tests of Within-Subjects Effects. Of the four Multivariate test results given, Wilks’ Lambda is commonly used. Note that the multivariate analysis yields significance at the 0.001 level for rest and at the 0.003 level for exercise. Since these are significant (p<.05), a univariate analysis follows to determine which physiologic parameters were significantly affected by the treatment. If the significance level is greater than 0.05, no further analysis is required, although SPSS will typically still complete the univariate analyses. Ignore the univariate results and report the F value and significance (e.g., F = 4.132; p = .204).
Multivariate analysis of resting data:
Multivariate analysis of exercise data:
According to statisticians, large sample numbers are required for multivariate analysis. We typically have small numbers of subjects in exercise physiology research. Consequently, we should just report the univariate results and specify the dependent variables that were measured (e.g., VO2, HR, BP, etc.) so the reader clearly knows how many variables were statistically analyzed.
The univariate analysis revealed that both HR and SBP are significantly affected by massage at rest (p<.001 for both HR and SBP) and during exercise (p<.002 for HR; p<.003 or SBP) following the massage. The F value for HR at rest is 117.042, for SBP, 152.111. For the exercise data, the F value for HR is 58.517, for SBP, 42.976. These data are summarized in the Tests of Within-Subjects Contrasts.
Resting data:
Exercise data:
Use a table to present your results in your research report (in the results section). Typically, the mean ± one standard deviation, the F value and the significance are reported for each dependent variable. The mean and standard deviation can be obtained by clicking on Analyze > Descriptive Statistics > Descriptives. Move all the dependent variables into the right window in the order in which you want them presented and click OK (see table below). Mean squares and/or sum of squares and degrees of freedom may be reported instead of the mean ± one standard deviation. Standard error of the estimate is sometimes reported in place of the standard deviation. The tabled results for example 1 are provided below. Remember, HR and SBP are reported in whole numbers.
Table 1. Physiologic responses comparing massage to no massage during rest and exercise.
Physiological variables / No Massage(Mean ± SD) / Massage
(Mean ± SD) / F value / Significance
HR, rest / 69 ± 4 / 58 ± 3 / 117.042 / <.001
HR, exercise / 158 ± 6 / 139 ± 4 / 58.517 / .002
SBP, rest / 125 ± 4 / 103 ± 2 / 152.111 / <.001
SBP, exercise / 160 ± 4 / 145 ± 3 / 42.976 / .003
We can report that a multivariate repeated measures ANOVA was used to evaluate the effects of massage at rest and during exercise. A significant effect was found for rest (Lambda(2,3) = 133.25, p = .001) and exercise (Lambda(2,3) = 76.75, p = .003). Follow-up univariate ANOVAs indicated that both HR and SBP are significantly affected by massage at rest (HR: F(1,4) = 117.04, p<.001; SBP: F(1,4) = 152.11, p<.001) and during exercise (HR: F(1,4) = 58.52, p = .002; SBP: F(1,4) = 42.98, p = .003).
Example 2
Check the first table of Within-Subjects Factors to verify that each dependent variable is identified correctly. Note that in the first column the upright position is identified as #1, supine as #2 and head down as #3. This should correlate with your identification scheme in the Dependent Variable column (e.g., SV1 = SVU, SV2 = SVS; SBP3 = SBPD, etc.).
Again, we are only interested in the Tests of Within-Subjects Effects. Of the four multivariate test results given, Wilks’ Lambda is preferred, which yields significance (p<.001) in this example.
Follow-up univariate ANOVAs are performed (i.e., Tests of Between Subjects Effects). The POSITION row shows us that SV is significant (p<.001), but SBP is not (p = .874). We are using the Sphericity Assumed values.
Since SV is significantly affected by body position, we have to determine which body positions differ significantly for SV. To do this, we will perform a paired t-test that compares the mean SV of each body position. Go back to the Data View, click on Analyze > Compare Means > Paired Samples T Test. Highlight each pair of variables that need to be compared (svu – svs, svu – svd, svs-svd). You should have three paired comparisons in the window under Paired Variables (see figure below). Click on OK. Remember to save your output data to a floppy disk.
The t value and significance are provided in the Paired Samples Test table. Notice that you also have the means and standard deviations for SVs in each position in the first table (Paired Samples Statistics). To get the means and standard deviations for SBP, you can click Analyze > Descriptive Statistics > Descriptives. Move the SBP variables into the right window in the order in which you want them presented and click OK.
Since we used each group twice in our comparisons, we have to perform a Bonferroni correction to control for a Type I error. To do this, we divide our original alpha value (.05) by the number of paired comparisons we made (3). Our new alpha value is .017 (.05 ÷ 3). The significance for each comparison is < .017, so we can report that SV differs significantly between upright and supine, upright and head down, and supine and head down positions. As in example 1, present the results in a table and include the mean ± one standard deviation, the t value and the significance for each dependent variable.
Table 2. Physiologic responses comparing upright, supine, and head down positions.
Physiological variables / Mean ± SDA / Mean ± SD
B / Mean ± SD
C / t value / Significance
SV / 69 ± 3
A-B
A-C / 75 ± 3
B-C / 79 ± 3 / -10.251
6.517
-8.773 / .001*
.003*
.001*
SBP / 115 ± 4 / 114 ± 3 / 115 ± 5 / # / #
*p<.05
#t test not done since the multivariate analysis revealed no significance
A = upright
B = supine
C = head down
We can report that we compared SV and SBP in three body positions (upright, supine, head down) using a multivariate repeated measures ANOVA. A significant effect was found (Lambda(4,14) = 13.34, p<.001). Follow-up univariate ANOVAs indicated that SBP was not significantly affected by body position (F(2,8) = .137, p = .874), but SV was (F(2,8) = 77.25, p<.001). Follow-up paired t tests revealed that SV significantly increased from upright to supine (t = -10.25, p = .001), supine to head down (t = 6.52, p = .003), and upright to head down positions (t = -8.77, p = .001).
Hopefully, these examples will enable you to use SPSS to analyze your data correctly. If you have any questions, check with the instructor before proceeding.
One more type of statistical analysis should be described just in case someone actually finds enough subjects to have two separate groups for the control and treatment (i.e., not repeated measures). In this case, the appropriate statistical analysis would be a MANOVA (multivariate ANOVA). The MANOVA does not involve repeated measures (different subjects in each group), but does incorporate multiple dependent variables.
MANOVA
Let’s repeat the tilt table experiment, but use 15 subjects, 5 in the upright position, 5 supine, and 5 head down. In order to perform a MANOVA in SPSS, we will have to rearrange our data somewhat. Use the Variable View to describe each dependent variable (e.g., sv, sbp). Also describe the independent variable (treatment) by naming it “position” and setting the decimal setting to 0. You now have 3 columns in the Data View. Arrange your data such that the first 5 cells (1-5) in the sv and sbp columns contain the SV and SBP values for the subjects in the upright position. Assign the number 0 to these five subjects in the position column. Put the supine data in the next 5 cells (6-10) and the head down data in the last 5 cells (11-15). Assign the number 1 in the position column to the supine values (cells 6-10) and number 2 to the head down values (cells 11-15). The Data View should appear as follows: