Research model drawer

How to draw a research model online

A research model (or conceptual framework diagram) shows the constructs you will measure and the causal paths you will test. Each arrow is one hypothesis. General drawing tools such as draw.io, Canva or Visio can draw the boxes, but they do not understand the model: they do not write the hypotheses, count arrows for the sample size or pass the model on to an analysis. This tool does all three and still exports an SVG you can polish in PowerPoint or draw.io.

There are four ways to start, and each produces the same tidy left-to-right diagram:

1. Paste hypotheses. Copy the hypotheses from your proposal, one per line or as a paragraph “H1: … H2: …”. The tool reads arrows (SQ -> SAT, SQ, PV, TR -> SAT -> LOY), sentences such as “Perceived risk negatively affects purchase intention”, “M mediates the relationship between X and Y” and “W moderates the relationship between X and Y”, prefixes like H2 (-):, and codes in brackets. Before drawing, it shows a table of what it understood, with a ✓ or ✗ for each line.
2. Upload data. Choose a .csv or .xlsx file or a Google Forms export. Only the header row is read, in your browser; columns that share a prefix (SAT1, SAT2, SAT3) become one construct, SAT. You then pick the dependent variable and any mediators.
3. Template models. One click loads TAM, TPB, UTAUT2, the expectation–confirmation model, SERVQUAL or SERVPERF leading to satisfaction and loyalty, or an ACSI-style customer satisfaction model, with cited scales and item counts from the scale library.
4. Model image. Put an existing picture under the canvas and trace over it. The tool can read the box labels with Tesseract, but you draw the arrows yourself.

Worked example: the TAM model

Choose “Template models”, then TAM (or click “Load the TAM example”). Davis’s (1989) Technology Acceptance Model has three constructs, Perceived Ease of Use (PEOU), Perceived Usefulness (PU) and Behavioral Intention to Use (BI), and three arrows: PEOU → PU, PU → BI and PEOU → BI. The tool writes:

• H1: Perceived Ease of Use (PEOU) has a positive effect on Perceived Usefulness (PU).
• H2: Perceived Usefulness (PU) has a positive effect on Behavioral Intention to Use (BI).
• H3: Perceived Ease of Use (PEOU) has a positive effect on Behavioral Intention to Use (BI).
• H4: Perceived Usefulness (PU) mediates the relationship between Perceived Ease of Use (PEOU) and Behavioral Intention to Use (BI).

BI receives the most arrows (2). G*Power (multiple regression, f² = 0.15, α = .05, power = .80, 2 predictors) gives 68 responses; the inverse square root method of Kock and Hadaya (2018), with a minimum path coefficient of .20, gives 155. The minimum sample is the larger figure, 155 valid responses, and allowing 15% for unusable responses you should collect about 183.

How to read the right-hand column

Hypotheses are numbered in arrow order; use the ↑ button to reorder them and the +/− button to switch between a positive and a negative effect. Minimum sample size counts the largest number of predictors pointing at any one construct, because in PLS-SEM each endogenous construct is estimated by a regression on the constructs that point to it (Hair et al., 2022). The figure is the larger of G*Power 3.1 (Faul et al., 2009) at a medium effect, f² = 0.15 (Cohen, 1992), and the inverse square root method. Needs attention flags loops, unconnected constructs and scales with fewer than 3 items.

Extended models

• Moderators. Drag the dot of box W onto the middle of arrow X → Y. W and the interaction term X×W are added as predictors of Y. Interaction effects are usually small: f² of about 0.005–0.035, with a mean of only 0.009 in the 30-year review by Aguinis et al. (2005). Tick the f² = 0.02 option to size the sample for the interaction itself, and estimate it with the two-stage approach (Becker, Ringle and Sarstedt, 2018).
• Mediators. Any X → M → Y chain gets a mediation hypothesis. Test the indirect effect with bootstrapping rather than the Baron and Kenny steps (Zhao et al., 2010). Fritz and MacKinnon (2007) show that when paths a and b are both small (.14) you need about 460–560 responses for power .80; when both are medium (.39), about 70–80.
• Higher-order constructs. Select several boxes and click “Higher-order”. A reflective–formative construct counts its components as predictors; when an arrow points into it, use the two-stage approach (Becker et al., 2012; Sarstedt et al., 2019).
• Multigroup analysis. Each group must reach the minimum sample on its own. Check measurement invariance with MICOM (Henseler, Ringle and Sarstedt, 2016) before comparing paths.
• Control variables. Dashed boxes with no hypothesis; each control arrow still counts as a predictor.

Reporting the model in APA 7 style

Download the PNG (at least 2,400 pixels wide, sharp at 300 dpi) and insert it at text width. Number and title the figure above it, for example “Figure 1. Proposed Research Model”, and add a note below when it builds on a published model: “Note. Adapted from Davis (1989).” Use the same construct abbreviations in the figure, the measurement table and the results tables.

FAQ

Is this research model drawer free?

Yes. No account is needed and the model stays in your browser; the “Link” button copies a URL that reopens it.

Can I edit the diagram in PowerPoint, Visio or draw.io?

Download the SVG and open or insert it there; every box, arrow and label stays editable.

What is the difference between a conceptual framework and a research model?

Many supervisors use the terms interchangeably. Strictly, the conceptual framework is the theoretical argument; the research model is its diagram of constructs and hypothesised paths, built on a base theory such as TAM or TPB plus the constructs you add and justify from prior studies.

When should a hypothesis be negative?

When theory and prior studies say that as one construct rises the other falls, for example perceived risk lowering purchase intention. The sign must match the coefficient you expect; the hypothesis is not supported when the coefficient is non-significant or has the opposite sign.

Do control variables need hypotheses?

No. Report their coefficients next to the hypotheses without concluding “supported” or “not supported”.